Chapter 7: The Earth’s Energy Budget, Climate Feedbacks, and Climate Sensitivity

Coordinating Lead Authors:

Piers Forster (United Kingdom), Trude Storelvmo (Norway)

Lead Authors:

Kyle Armour (United States of America), William Collins (United Kingdom), Jean-Louis Dufresne (France), David Frame (New Zealand), Daniel J. Lunt (United Kingdom), Thorsten Mauritsen (Sweden/Denmark), Matthew D. Palmer (United Kingdom), Masahiro Watanabe (Japan), Martin Wild (Switzerland), Hua Zhang (China)

Contributing Authors:

Kari Alterskjær (Norway), Chris Smith (United Kingdom), Govindasamy Bala (India/United States of America), Nicolas Bellouin (United Kingdom/France), Terje Berntsen (Norway), Fábio Boeira Dias (Finland/Brazil), Sandrine Bony (France), Natalie J. Burls (United States of America/South Africa), Michelle Cain (United Kingdom), Catia M. Domingues (Australia, United Kingdom/Brazil), Aaron Donohoe (United States of America), Mark Flanner (United States of America), Jan S. Fuglestvedt (Norway), Lily C. Hahn (United States of America), Glen R. Harris (United Kingdom/New Zealand, United Kingdom), Christopher Jones (United Kingdom), Seiji Kato (United States of America), Jared Lewis (Australia/New Zealand), Zhanqing Li (United States of America), Mike Lockwood (United Kingdom), Norman Loeb (United States of America), Jochem Marotzke (Germany), Malte Meinshausen (Australia/Germany), Sebastian Milinski (Germany), Zebedee R.J. Nicholls (Australia), Ryan S. Padron Flasher (Switzerland/Ecuador, United States of America), Anna Possner (Germany), Cristian Proistosescu (Romania), Johannes Quaas (Germany), Joeri Rogelj (United Kingdom/Belgium), Daniel Rosenfeld (Israel), Bjørn H. Samset (Norway), Abhishek Savita (Australia/India), Jessica Vial (France), Karina von Schuckmann (France/Germany), Mark Zelinka (United States of America), Shuyun Zhao (China)

Review Editors:

Robert Colman (Australia), H. Damon Matthews (Canada), Venkatachalam Ramaswamy (United States of America)

Chapter Scientists:

Kari Alterskjær (Norway), Chris Smith (United Kingdom)

Figure 7.1a

Figure 7.1b

Figure 7.2

Figure 7.3

Figure 7.4

Figure 7.5

Figure 7.6

Figure 7.7

Figure 7.8

Figure 7.9

Figure 7.10

Figure 7.11

Figure 7.12

Figure 7.13

Figure 7.14

Figure 7.15

Figure 7.16

Figure 7.17

Figure 7.18

Figure 7.19

Figure 7.20

Figure 7.21

Figure 7.22

Box 7.1, Figure 1

Box 7.2, Figure 1

Cross-Chapter Box 7.1, Figure 1

FAQ 7.1 Figure 1

FAQ 7.2, Figure 1

FAQ 7.3, Figure 1

This chapter should be cited as:

Forster, P., T. Storelvmo, K. Armour, W. Collins, J.-L. Dufresne, D. Frame, D.J. Lunt, T. Mauritsen, M.D. Palmer, M. Watanabe, M. Wild, and H. Zhang, 2021: The Earth’s Energy Budget, Climate Feedbacks, and Climate Sensitivity. In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Masson-Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R. Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, and B. Zhou (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 923–1054, doi: 10.1017/9781009157896.009.

Executive Summary

This chapter assesses the present state of knowledge of Earth’s energy budget: that is, the main flows of energy into and out of the Earth system, and how these energy flows govern the climate response to a radiative forcing. Changes in atmospheric composition and land use, like those caused by anthropogenic greenhouse gas emissions and emissions of aerosols and their precursors, affect climate through perturbations to Earth’s top-of-atmosphere energy budget. The effective radiative forcings (ERFs) quantify these perturbations, including any consequent adjustment to the climate system (but excluding surface temperature response). How the climate system responds to a given forcing is determined by climate feedbacks associated with physical, biogeophysical and biogeochemical processes. These feedback processes are assessed, as are useful measures of global climate response, namely equilibrium climate sensitivity (ECS) and the transient climate response (TCR). This chapter also assesses emissions metrics, which are used to quantify how the climate response to the emissions of different greenhouse gases compares to the response to the emissions of carbon dioxide (CO2). This chapter builds on the assessment of carbon cycle and aerosol processes from Chapters 5 and 6, respectively, to quantify non-CO2 biogeochemical feedbacks and the ERF for aerosols. Other chapters in this Report use this chapter’s assessment of ERF, ECS and TCR to help understand historical and future temperature changes (Chapters 3 and 4, respectively), the response to cumulative emissions and the remaining carbon budget (Chapter 5), emissions-based radiative forcing (Chapter 6) and sea level rise (Chapter 9). This chapter builds on findings from the IPCC Fifth Assessment Report (AR5), the Special Report on Global Warming of 1.5°C (SR1.5), the Special Report on the Ocean and Cryosphere in a Changing Climate (SROCC) and the Special Report on climate change, desertification, land degradation, sustainable land management, food security, and greenhouse gas luxes in terrestrial ecosystems (SRCCL). Very likely ranges are presented unless otherwise indicated.

Earth’s Energy Budget

Since AR5, the accumulation of energy in the Earth system, quantified by changes in the global energy inventory for all components of the climate system, has become established as a robust measure of the rate of global climate change on interannual-to-decadal time scales. Compared to changes in global surface air temperature (GSAT), the global energy inventory exhibits less variability, which can mask underlying climate trends. Compared to AR5, there is increased confidence in the quantification of changes in the global energy inventory due to improved observational records and closure of the sea level budget. Energy will continue to accumulate in the Earth system until at least the end of the 21st century, even under strong mitigation scenarios, and will primarily be observed through ocean warming and associated with continued sea level rise through thermal expansion (high confidence). {7.2.2, Box 7.2, Table 7.1, Cross-Chapter Box 9.1, Table 9.5, 9.2.2, 9.6.3}

The global energy inventory increased by 282 [177 to 387] Zettajoules (ZJ; 1021Joules) for the period 1971–2006 and 152 [100 to 205] ZJ for the period 2006–2018. This corresponds to an Earth energy imbalance of 0.50 [0.32 to 0.69] W m–2 for the period 1971–2006, increasing to 0.79 [0.52 to 1.06] W m–2 for the period 2006–2018, expressed per unit area of Earth’s surface. Ocean heat uptake is by far the largest contribution and accounts for 91% of the total energy change. Compared to AR5, the contribution from land heating has been revised upwards from about 3% to about 5%. Melting of ice and warming of the atmosphere account for about 3% and 1% of the total change respectively. More comprehensive analysis of inventory components and cross-validation of global heating rates from satellite and in situ observations lead to a strengthened assessment relative to AR5 (high confidence). {Box 7.2, 7.2.2, Table 7.1, 7.5.2.3}

Improved quantification of effective radiative forcing, the climate system radiative response, and the observed energy increase in the Earth system for the period 1971–2018 demonstrate improved closure of the global energy budget compared to AR5. Combining the likely range of ERF with the central estimate of radiative response gives an expected energy gain of 340 [47 to 662] ZJ. Combining the likely range of climate response with the central estimate of ERF gives an expected energy gain of 340 [147 to 527] ZJ. Both estimates are consistent with an independent observation-based assessment of the global energy increase of 284 [96 to 471] ZJ, (very likely range) expressed relative to the estimated 1850–1900 Earth energy imbalance (high confidence). {7.2.2, Box 7.2, 7.3.5, 7.5.2}

Since AR5, additional evidence for a widespread decline (or dimming) in solar radiation reaching the surface is found in the observational records between the 1950s and 1980s, with a partial recovery (brightening) at many observational sites thereafter (high confidence). These trends are neither a local phenomenon nor a measurement artefact (high confidence). Multi-decadal variation in anthropogenic aerosol emissions are thought to be a major contributor (medium confidence), but multi-decadal variability in cloudiness may also have played a role. The downward and upward thermal radiation at the surface has increased in recent decades, in line with increased greenhouse gas concentrations and associated surface and atmospheric warming and moistening (medium confidence). {7.2.2}

Effective Radiative Forcing

For carbon dioxide, methane, nitrous oxide and chlorofluorocarbons, there is now evidence to quantify the effect on ERF of tropospheric adjustments (e.g., from changes in atmospheric temperatures, clouds and water vapour). The assessed ERF for a doubling of carbon dioxide compared to 1750 levels (3.93 ± 0.47 W m–2) is larger than in AR5. Effective radiative forcings (ERF), introduced in AR5, have been estimated for a larger number of agents and shown to be more closely related to the temperature response than the stratospheric-temperature adjusted radiative forcing. For carbon dioxide, the adjustments include the physiological effects on vegetation (high confidence). {7.3.2}

The total anthropogenic ERF over the industrial era (1750–2019) was 2.72 [1.96 to 3.48] W m–2. This estimate has increased by 0.43 W m–2compared to AR5 estimates for 1750–2011. This increase includes +0.34 W m–2 from increases in atmospheric concentrations of well-mixed greenhouse gases (including halogenated species) since 2011, +0.15 W m–2 from upwards revisions of their radiative efficiencies and +0.10 W m–2 from re-evaluation of the ozone and stratospheric water vapour ERF. The 0.59 W m–2 increase in ERF from greenhouse gases is partly offset by a better-constrained assessment of total aerosol ERF that is more strongly negative than in AR5, based on multiple lines of evidence (high confidence). Changes in surface reflectance from land-use change, deposition of light-absorbing particles on ice and snow, and contrails and aviation-induced cirrus have also contributed to the total anthropogenic ERF over the industrial era, with –0.20 [–0.30 to –0.10] W m–2(medium confidence), +0.08 [0 to 0.18] W m–2(low confidence) and +0.06 [0.02 to 0.10] W m–2(low confidence), respectively. {7.3.2, 7.3.4, 7.3.5}

Anthropogenic emissions of greenhouse gases and their precursors contribute an ERF of 3.84 [3.46 to 4.22] W m–2 over the industrial era (1750–2019). Most of this total ERF, 3.32 [3.03 to 3.61] W m–2, comes from the well-mixed greenhouse gases, with changes in ozone and stratospheric water vapour (from methane oxidation) contributing the remainder. The ERF of greenhouse gases is composed of 2.16 [1.90 to 2.41] W m–2 from carbon dioxide, 0.54 [0.43 to 0.65] W m–2 from methane, 0.41 [0.33 to 0.49] W m–2 from halogenated species, and 0.21 [0.18 to 0.24] W m–2 from nitrous oxide. The ERF for ozone is 0.47 [0.24 to 0.71] W m–2. The estimate of ERF for ozone has increased since AR5 due to revised estimates of precursor emissions and better accounting for effects of tropospheric ozone precursors in the stratosphere. The estimated ERF for methane has slightly increased due to a combination of increases from improved spectroscopic treatments being somewhat offset by accounting for adjustments (high confidence). {7.3.2, 7.3.5}

Aerosols contribute an ERF of –1.3 [–2.0 to –0.6] W m–2 over the industrial era (1750–2014) (medium confidence). The ERF due to aerosol–cloud interactions (ERFaci) contributes most to the magnitude of the total aerosol ERF (high confidence) and is assessed to be –1.0 [–1.7 to –0.3] W m–2 ( medium confidence), with the remainder due to aerosol–radiation interactions (ERFari), assessed to be –0.3 [–0.6 to 0.0] W m–2 ( medium confidence). There has been an increase in the estimated magnitude but a reduction in the uncertainty of the total aerosol ERF relative to AR5, supported by a combination of increased process-understanding and progress in modelling and observational analyses. ERF estimates from these separate lines of evidence are now consistent with each other, in contrast to AR5, and support the assessment that it is virtually certain that the total aerosol ERF is negative. Compared to AR5, the assessed magnitude of ERFaci has increased, while the magnitude of ERFari has decreased . The total aerosol ERF over the period 1750–2019 is less certain than the headline statement assessment. It is also assessed to be smaller in magnitude at –1.1 [–1.7 to –0.4] W m–2, primarily due to recent emissions changes (medium confidence). {7.3.3, 7.3.5, 2.2.6}

Climate Feedbacks and Sensitivity

The net effect of changes in clouds in response to global warming is to amplify human-induced warming, that is, the net cloud feedback is positive (high confidence). Compared to AR5, major advances in the understanding of cloud processes have increased the level of confidence and decreased the uncertainty range in the cloud feedback by about 50%. An assessment of the low-altitude cloud feedback over the subtropical oceans, which was previously the major source of uncertainty in the net cloud feedback, is improved owing to a combined use of climate model simulations, satellite observations, and explicit simulations of clouds, altogether leading to strong evidence that this type of cloud amplifies global warming. The net cloud feedback, obtained by summing the cloud feedbacks assessed for individual regimes, is 0.42 [–0.10 to +0.94] W m–2°C–1. A net negative cloud feedback is very unlikely (high confidence). {7.4.2, Figure 7.10, Table 7.10}

The combined effect of all known radiative feedbacks (physical, biogeophysical, and non-CO2 biogeochemical) is to amplify the base climate response, also known as the Planck temperature response (virtually certain). Combining these feedbacks with the base climate response, the net feedback parameter based on process understanding is assessed to be –1.16 [–1.81 to –0.51] W m–2°C–1, which is slightly less negative than that inferred from the overall ECS assessment. The combined water-vapour and lapse-rate feedback makes the largest single contribution to global warming, whereas the cloud feedback remains the largest contribution to overall uncertainty. Due to the state-dependence of feedbacks, as evidenced from paleoclimate observations and from models, the net feedback parameter will increase (become less negative) as global temperature increases. Furthermore, on long time scales the ice-sheet feedback parameter is very likely positive, promoting additional warming on millennial time scales as ice sheets come into equilibrium with the forcing (high confidence). {7.4.2, 7.4.3, 7.5.7}

Radiative feedbacks, particularly from clouds, are expected to become less negative (more amplifying) on multi-decadal time scales as the spatial pattern of surface warming evolves, leading to an ECS that is higher than was inferred in AR5 based on warming over the instrumental record. This new understanding, along with updated estimates of historical temperature change, ERF, and Earth’s energy imbalance, reconciles previously disparate ECS estimates (high confidence). However, there is currently insufficient evidence to quantify a likely range of the magnitude of future changes to current climate feedbacks. Warming over the instrumental record provides robust constraints on the lower end of the ECS range (high confidence), but owing to the possibility of future feedback changes it does not, on its own, constrain the upper end of the range, in contrast to what was reported in AR5. {7.4.4, 7.5.2, 7.5.3}

Based on multiple lines of evidence the best estimate of ECS is 3°C, the likely range is 2.5°C to 4°C, and the very likely range is 2°C to 5°C. It is virtually certain that ECS is larger than 1.5°C. Substantial advances since AR5 have been made in quantifying ECS based on feedback process understanding, the instrumental record, paleoclimates and emergent constraints. There is a high level of agreement among the different lines of evidence. All lines of evidence help rule out ECS values below 1.5°C, but currently it is not possible to rule out ECS values above 5°C. Therefore, the 5°C upper end of the very likely range is assessed to have medium confidence and the other bounds have high confidence. {7.5.5}

Based on process understanding, warming over the instrumental record, and emergent constraints, the best estimate of TCR is 1.8°C, the likely range is 1.4°C to 2.2°C and the very likely range is 1.2°C to 2.4°C (high confidence). {7.5.5}

On average, Coupled Model Intercomparison Project Phase 6 (CMIP6) models have higher mean ECS and TCR values than the Phase 5 (CMIP5) generation of models. They also have higher mean values and wider spreads than the assessed best estimates and very likely ranges within this Report. These higher ECS and TCR values can, in some models, be traced to changes in extra-tropical cloud feedbacks that have emerged from efforts to reduce biases in these clouds compared to satellite observations (medium confidence). The broader ECS and TCR ranges from CMIP6 also lead the models to project a range of future warming that is wider than the assessed warming range, which is based on multiple lines of evidence. However, some of the high-sensitivity CMIP6 models are less consistent with observed recent changes in global warming and with paleoclimate proxy data than models with ECS within the very likely range. Similarly, some of the low-sensitivity models are less consistent with the paleoclimate data. The CMIP models with the highest ECS and TCR values provide insights into low-likelihood, high-impact outcomes, which cannot be excluded based on currently available evidence (high confidence). {4.3.1, 4.3.4, 7.4.2, 7.5.6}

Climate Response

The total human-forced GSAT change from 1750 to 2019 is calculated to be 1.29 [0.99 to 1.65] °C. This calculation is an emulator-based estimate, constrained by the historic GSAT and ocean heat content changes from (Chapter 2 and the ERF, ECS and TCR from this chapter. The calculated GSAT change is composed of a well-mixed greenhouse gas warming of 1.58 [1.17 to 2.17] °C (high confidence), a warming from ozone changes of 0.23 [0.11 to 0.39] °C (high confidence), a cooling of –0.50 [–0.22 to –0.96] °C from aerosol effects (medium confidence), and a –0.06 [–0.15 to +0.01] °C contribution from surface reflectance changes from land-use change and light-absorbing particles on ice and snow (medium confidence). Changes in solar and volcanic activity are assessed to have together contributed a small change of –0.02 [–0.06 to +0.02] °C since 1750 (medium confidence). {7.3.5}

Uncertainties regarding the true value of ECS and TCR are the dominant source of uncertainty in global temperature projections over the 21st century under moderate to high greenhouse gas emissions scenarios. For scenarios that reach net zero carbon dioxide emissions, the uncertainty in the ERF values of aerosol and other short-lived climate forcers contribute substantial uncertainty in projected temperature. Global ocean heat uptake is a smaller source of uncertainty in centennial-time scale surface warming (high confidence). {7.5.7}

The assessed historical and future ranges of GSAT change in this Report are shown to be internally consistent with the Report’s assessment of key physical-climate indicators: greenhouse gas ERFs, ECS and TCR. When calibrated to match the assessed ranges within the assessment, physically based emulators can reproduce the best estimate of GSAT change over 1850–1900 to 1995–2014 to within 5% and the very likely range of this GSAT change to within 10%. Two physically based emulators match at least two-thirds of the Chapter 4-assessed projected GSAT changes to within these levels of precision. When used for multi-scenario experiments, calibrated physically based emulators can adequately reflect assessments regarding future GSAT from Earth system models and/or other lines of evidence (high confidence). {Cross-Chapter Box 7.1}

It is now well understood that the Arctic warms more quickly than the Antarctic due to differences in radiative feedbacks and ocean heat uptake between the poles, but that surface warming will eventually be amplified in both the Arctic and Antarctic (high confidence). The causes of this polar amplification are well understood, and the evidence is stronger than at the time of AR5, supported by better agreement between modelled and observed polar amplification during warm paleo time periods (high confidence). The Antarctic warms more slowly than the Arctic owing primarily to upwelling in the Southern Ocean, and even at equilibrium is expected to warm less than the Arctic. The rate of Arctic surface warming will continue to exceed the global average over this century (high confidence). There is also high confidence that Antarctic amplification will emerge as the Southern Ocean surface warms on centennial time scales, although onlylow confidence regarding whether this feature will emerge during the 21st century. {7.4.4}

The assessed global warming potentials (GWP) and global temperature-change potentials (GTP) for methane and nitrous oxide are slightly lower than in AR5 due to revised estimates of their lifetimes and updated estimates of their indirect chemical effects (medium confidence). The assessed metrics now also include the carbon cycle response for non-CO2 gases. The carbon cycle estimate is lower than in AR5, but there is high confidence in the need for its inclusion and in the quantification methodology. Metrics for methane from fossil fuel sources account for the extra fossil CO2 that these emissions contribute to the atmosphere and so have slightly higher emissions metric values than those from biogenic sources (high confidence). {7.6.1}

New emissions metric approaches such as GWP* and the combined-GTP (CGTP) are designed to relate emissions rates of short-lived gases to cumulative emissions of CO2 . These metric approaches are well suited to estimate the GSAT response from aggregated emissions of a range of gases over time, which can be done by scaling the cumulative CO2 equivalent emissions calculated with these metrics by the transient climate response to cumulative emissions of CO2 . For a given multi-gas emissions pathway, the estimated contribution of emissions to surface warming is improved by using either these new metric approaches or by treating short- and long-lived GHG emissions pathways separately, as compared to approaches that aggregate emissions of GHGs using standard GWP or GTP emissions metrics. By contrast, if emissions are weighted by their 100-year GWP or GTP values, different multi-gas emissions pathways with the same aggregated CO2 equivalent emissions rarely lead to the same estimated temperature outcome (high confidence). {7.6.1, Box 7.3}

The choice of emissions metric affects the quantification of net zero GHG emissions and therefore the resulting temperature outcome after net zero emissions are achieved. In general, achieving net zero CO2 emissions and declining non-CO2 radiative forcing would be sufficient to prevent additional human-caused warming. Reaching net zero GHG emissions as quantified by GWP-100 typically results in global temperatures that peak and then decline after net zero GHGs emissions are achieved, though this outcome depends on the relative sequencing of mitigation of short-lived and long-lived species. In contrast, reaching net zero GHG emissions when quantified using new emissions metrics such as CGTP or GWP* would lead to approximate temperature stabilization (high confidence). {7.6.2}

7.1 Introduction, Conceptual Framework, and Advances Since the Fifth Assessment Report

This chapter assesses the major physical processes that affect the evolution of Earth’s energy budget and the associated changes in surface temperature and the broader climate system, integrating elements that were dealt with separately in previous reports.

The top-of-atmosphere (TOA) energy budget determines the net amount of energy entering or leaving the climate system. Its time variations can be monitored in three ways, using: (i) satellite observations of the radiative fluxes at the TOA; (ii) observations of the accumulation of energy in the climate system; and (iii) observations of surface energy fluxes. When the TOA energy budget is changed by a human or natural cause (a ‘radiative forcing’), the climate system responds by warming or cooling (i.e., the system gains or loses energy). Understanding of changes in the Earth’s energy flows helps understanding of the main physical processes driving climate change. It also provides a fundamental test of climate models and their projections.

This chapter principally builds on the IPCC Fifth Assessment Report (AR5; Boucher, 2012; Church et al., 2013; M. Collins et al., 2013; Flato et al., 2013; Hartmann et al., 2013; Myhre et al., 2013b; Rhein et al., 2013). It also builds on the subsequent IPCC Special Report on Global Warming of 1.5°C (SR1.5; IPCC, 2018), the Special Report on the Ocean and Cryosphere in a Changing Climate (SROCC; IPCC, 2019a) and the Special Report on climate change, desertification, land degradation, sustainable land management, food security, and greenhouse gas fluxes in terrestrial ecosystems (SRCCL; IPCC, 2019b), as well as community-led assessments (e.g., Bellouin et al. (2020) covering aerosol radiative forcing and Sherwood et al. (2020) covering equilibrium climate sensitivity).

Throughout this chapter, global surface air temperature (GSAT) is used to quantify surface temperature change (Cross-Chapter Box 2.3 and (Section 4.3.4). The total energy accumulation in the Earth system represents a metric of global change that is complementary to GSAT but shows considerably less variability on interannual-to-decadal time scales (Section 7.2.2). Research and new observations since AR5 have improved scientific confidence in the quantification of changes in the global energy inventory and corresponding estimates of Earth’s energy imbalance (Section 7.2). Improved understanding of adjustments to radiative forcing and of aerosol–cloud interactions have led to revisions of forcing estimates (Section 7.3). New approaches to the quantification and treatment of feedbacks (Section 7.4) have improved the understanding of their nature and time-evolution, leading to a better understanding of how these feedbacks relate to equilibrium climate sensitivity (ECS). This has helped to reconcile disparate estimates of ECS from different lines of evidence (Section 7.5). Innovations in the use of emissions metrics have clarified the relationships between metric choice and temperature policy goals (Section 7.6), linking this chapter to WGIII which provides further information on metrics, their use, and policy goals beyond temperature. Very likely (5–95%) ranges are presented unless otherwise indicated. In particular, the addition of ‘(one standard deviation)’ indicates that the range represents one standard deviation.

In Box 7.1 an energy budget framework is introduced, which forms the basis for the discussions and scientific assessment in the remainder of this chapter and across the Report. The framework reflects advances in the understanding of the Earth system response to climate forcing since the publication of AR5. A schematic of this framework and the key changes relative to the science reported in AR5 are provided in Figure 7.1.

Figure 7.1 | Visual guide to Chapter 7. Panel (a) Overview of the chapter.

Figure 7.1: Panel (b) Visual abstract of the chapter, illustrating why the Earth’s energy budget matters and how it relates to the underlying chapter assessment. The methods used to assess processes and key new findings relative to AR5 are highlighted. Upper schematic adapted from Von Schuckmann et al. (2020).

A simple way to characterize the behaviour of multiple aspects of the climate system at once is to summarize them using global-scale metrics. This Report distinguishes between ‘climate metrics’ (e.g., ECS, TCR) and ‘emissions metrics’ (e.g., global warming potential, GWP, or global temperature-change potential, GTP), but this distinction is not definitive. Climate metrics are generally used to summarize aspects of the surface temperature response (Box 7.1). Emissions metrics are generally used to summarize the relative effects of emissions of different forcing agents, usually greenhouse gases (GHGs; Section 7.6). The climate metrics used in this report typically evaluate how the Earth system response varies with atmospheric gas concentration or change in radiative forcing. Emissions metrics evaluate how radiative forcing or a key climate variable (such as GSAT) is affected by the emissions of a certain amount of gas. Emissions-related metrics are sometimes used in mitigation policy decisions such as trading GHG reduction measures and life cycle analysis. Climate metrics are useful to gauge the range of future climate impacts for adaptation decisions under a given emissions pathway. Metrics such as the transient climate response to cumulative emissions of carbon dioxide (TCRE) are used in both adaptation and mitigation contexts: for gauging future global surface temperature change under specific emissions scenarios, and to estimate remaining carbon budgets that are used to inform mitigation policies (Section 5.5).

Given that TCR and ECS are metrics of GSAT response to a theoretical doubling of atmospheric CO2 (Box 7.1), they do not directly correspond to the warming that would occur under realistic forcing scenarios that include time-varying CO2 concentrations and non-CO2 forcing agents (such as aerosols and land-use changes). It has been argued that TCR, as a metric of transient warming, is more policy-relevant than ECS (Frame et al., 2006; Schwartz, 2018). However, as detailed in Chapter 4, both established and recent results (Forster et al., 2013; Gregory et al., 2015; Marotzke and Forster, 2015; Grose et al., 2018; Marotzke, 2019) indicate that TCR and ECS help explain variation across climate models both over the historical period and across a range of concentration-driven future scenarios. In emission-driven scenarios the carbon cycle response is also important (Smith et al., 2019). The proportion of variation explained by ECS and TCR varies with scenario and the time period considered, but both past and future surface warming depend on these metrics (Section 7.5.7).

Regional changes in temperature, rainfall, and climate extremes have been found to correlate well with the forced changes in GSAT within Earth System Models (ESMs; Section 4.6.1; Giorgetta et al., 2013; Tebaldi and Arblaster, 2014; Seneviratne et al., 2016). While this so-called ‘pattern scaling’ has important limitations arising from, for instance, localized forcings, land-use changes, or internal climate variability (Deser et al., 2012; Luyssaert et al., 2014), changes in GSAT nonetheless explain a substantial fraction of inter-model differences in projections of regional climate changes over the 21st century (Tebaldi and Knutti, 2018). This Chapter’s assessments of TCR and ECS thus provide constraints on future global and regional climate change (Chapters 4 and 11).

Box 7.1 | The Energy Budget Framework: Forcing and Response

The forcing and response energy budget framework provides a methodology to assess the effect of individual drivers of global surface temperature response, and to facilitate the understanding of the key phenomena that set the magnitude of this temperature response. The framework used here is developed from that adopted in previous IPCC reports (see Ramaswamy et al., 2019 for a discussion). Effective Radiative Forcing (ERF), introduced in AR5 (Boucher et al., 2013; Myhre et al., 2013b) is more explicitly defined in this Report and is employed as the central definition of radiative forcing (Sherwood et al., 2015, Box 7.1, Figure 1a). The framework has also been extended to allow variations in feedbacks over different time scales and with changing climate state (Sections 7.4.3 and 7.4.4).

The global surface air temperature (GSAT) response to perturbations that give rise to an energy imbalance is traditionally approximated by the following linear energy budget equation, in which ΔN represents the change in the top-of-atmosphere (TOA) net energy flux, ΔF is an effective radiative forcing perturbation to the TOA net energy flux, α is the net feedback parameter and ΔT is the change in GSAT:

ΔN =ΔF + α ΔT

ERF is the TOA energy budget change resulting from the perturbation, excluding any radiative response related to a change in GSAT (i.e., ΔT= 0). Climate feedbacks ( α ) represent those processes that change the TOA energy budget in response to a given ΔT.

The effective radiative forcing, ERFF; units: W m–2) quantifies the change in the net TOA energy flux of the Earth system due to an imposed perturbation (e.g., changes in greenhouse gas or aerosol concentrations, in incoming solar radiation, or land-use change). ERF is expressed as a change in net downward radiative flux at the TOA following adjustments in both tropospheric and stratospheric temperatures, water vapour, clouds, and some surface properties, such as surface albedo from vegetation changes, that are uncoupled to any GSAT change (Smith et al., 2018b). These adjustments affect the TOA energy balance and hence the ERF. They are generally assumed to be linear and additive (Section 7.3.1). Accounting for such processes gives an estimate of ERF that is more representative of the climate change response associated with forcing agents than stratospheric-temperature-adjusted radiative forcing (SARF) or the instantaneous radiative forcing (IRF; Section 7.3.1). Adjustments are processes that are independent of GSAT change, whereas feedbacks refer to processes caused by GSAT change. Although adjustments generally occur on time scales of hours to several months, and feedbacks respond to ocean surface temperature changes on time scales of a year or more, time scale is not used to separate the definitions. ERF has often been approximated as the TOA energy balance change due to an imposed perturbation in climate model simulations with sea surface temperature and sea-ice concentrations set to their pre-industrial climatological values (e.g., Forster et al., 2016). However, to match the adopted forcing–feedback framework, the small effects of any GSAT change from changes in land surface temperatures need to be removed from the TOA energy balance in such simulations to give an approximate measure of ERF (Box 7.1, Figure 1b and (Section 7.3.1).

Box 7.1, Figure 1 | Schematics of the forcing–feedback framework adopted within the assessment, following Equation 7.1. The figure illustrates how the Earth’s top-of-atmosphere (TOA) net energy flux might evolve for a hypothetical doubling of atmospheric CO2 concentration above pre-industrial levels, where an initial positive energy imbalance (energy entering the Earth system, shown on the y-axis) is gradually restored towards equilibrium as the surface temperature warms (shown on the x-axis). (a) illustrates the definitions of effective radiative forcing (ERF) for the special case of a doubling of atmospheric CO2 concentration, the feedback parameter and the equilibrium climate sensitivity (ECS). (b) illustrates how approximate estimates of these metrics are made within the chapter and how these approximations might relate to the exact definitions adopted in panel (a).

The feedback parameter, α (units: W m–2°C–1) quantifies the change in net energy flux at the TOA for a given change in GSAT. Many climate variables affect the TOA energy budget, and the feedback parameter can be decomposed, to first order, into a sum of terms

where x represents a variable of the Earth system that has a direct effect on the energy budget at the TOA. The sum of the feedback terms (i.e., α in Equation 7.1) governs Earth’s equilibrium GSAT response to an imposed ERF. In previous assessments, α and the related ECS have been associated with a distinct set of physical processes (Planck response and changes in water vapour, lapse rate, surface albedo, and clouds; Charney et al., 1979). In this assessment, a more general definition of α and ECS is adopted such that they include additional Earth system processes that act across many time scales (e.g., changes in natural aerosol emissions or vegetation). Because, in our assessment, these additional processes sum to a near-zero value, including these additional processes does not change the assessed central value of ECS but does affect its assessed uncertainty range (Section 7.4.2). Note that there is no standardized notation or sign convention for the feedback parameter in the literature. Here the convention is used that the sum of all feedback terms (the net feedback parameter, α ) is negative for a stable climate that radiates additional energy to space with a GSAT increase, with a more negative value of α corresponding to a stronger radiative response and thus a smaller GSAT change required to balance a change in ERF (Equation 7.1).

A change in process x amplifies the temperature response to a forcing when the associated feedback parameter α x is positive (positive feedback) and dampens the temperature response when α x is negative (negative feedback). New research since AR5 emphasizes how feedbacks can vary over different time scales (Section 7.4.4) and with climate state (Section 7.4.3), giving rise to the concept of an ‘effective feedback parameter’ that may be different from the equilibrium value of the feedback parameter governing ECS (Section 7.4.3).

The equilibrium climate sensitivity, ECS (units: °C), is defined as the equilibrium value of ΔT in response to a sustained doubling of atmospheric CO2 concentration from a pre-industrial reference state. The value of ERF for this scenario is denoted by ΔF2xCO2, giving ECS = –ΔF2xCO2/ α from Equation 7.1 applied at equilibrium (Box 7.1, Figure 1a and (Section 7.5). ‘Equilibrium’ refers to a steady state where ΔN averages to zero over a multi-century period. ECS is representative of the multi-century to millennial ΔTresponse to ΔF2xCO2, and is based on a CO2 concentration change so any feedbacks that affect the atmospheric concentration of CO2 do not influence its value. As employed here, ECS also excludes the long-term response of the ice sheets (Section 7.4.2.6) which may take multiple millennia to reach equilibrium, but includes all other feedbacks. Due to a number of factors, studies rarely estimate ECS or α at equilibrium or under CO2 forcing alone. Rather, they give an‘effective feedback parameter’ (Section 7.4.1 and Box 7.1, Figure 1b) or an ‘effective ECS’ (Section 7.5.1 and Box 7.1, Figure 1b), which represent approximations to the true values of α orECS. The ‘effective ECS’ represents the equilibrium value of ΔT in response to a sustained doubling of atmospheric CO2 concentration that would occur assuming the ‘effective feedback parameter’ applied at that equilibrium state. For example, a feedback parameter can be estimated from the linear slope of Δn against ΔTover a set number of years within ESM simulations of an abrupt doubling or quadrupling of atmospheric CO2 (2×CO2 or 4×CO2 , respectively), and the ECS can be estimated from the intersect of this regression line with ΔN = 0 (Box 7.1, Figure 1b). To infer ECS from a given estimate of effective ECS necessitates that assumptions are made for how ERF varies with CO2 concentration (Section 7.3.2) and how the slope of ΔN against Δ T relates to the slope of the straight line from ERF to ECS (Section 7.5 and Box 7.1, Figure 1b). Care has to be taken when comparing results across different lines of evidence to translate their estimates of the effective ECS into the ECS definition used here (Section 7.5.5).

The transient climate response, TCR (units: °C), is defined as the ΔT for the hypothetical scenario in which CO2 increases at 1% yr–1from a pre-industrial reference state to the time of a doubling of atmospheric CO2 concentration (year 70; Section 7.5). TCR is based on a CO2 concentration change, so any feedbacks that affect the atmospheric concentration of CO2 do not influence its value. It is a measure of transient warming accounting for the strength of climate feedbacks and ocean heat uptake. The transient climate response to cumulative emissions of carbon dioxide (TCRE) is defined as the transient ΔTper 1000 Gt C of cumulative CO2 emissions increase since the pre-industrial period. TCRE combines information on the airborne fraction of cumulative CO2 emissions (the fraction of the total CO2 emitted that remains in the atmosphere at the time of doubling, which is determined by carbon cycle processes) with information on the TCR. TCR is assessed in this chapter, whereas TCRE is assessed in (Chapter 5 (Section 5.5).

7.2 Earth’s Energy Budget and its Changes Through Time

Earth’s energy budget encompasses the major energy flows of relevance for the climate system (Figure 7.2). Virtually all the energy that enters or leaves the climate system does so in the form of radiation at the TOA. The TOA energy budget is determined by the amount of incoming solar (shortwave) radiation and the outgoing radiation that is composed of reflected solar radiation and outgoing thermal (longwave) radiation emitted by the climate system. In a steady-state climate, the outgoing and incoming radiative components are essentially in balance in the long-term global mean, although there are still fluctuations around this balanced state that arise through internal climate variability (Brown et al., 2014; Palmer and McNeall, 2014). However, anthropogenic forcing has given rise to a persistent imbalance in the global mean TOA radiation budget that is often referred to as Earth’s energy imbalance (e.g., Trenberth et al., 2014; von Schuckmann et al., 2016), which is a key element of the energy budget framework (N; Box 7.1, Equation 7.1) and an important metric of the rate of global climate change (Hansen et al., 2005a; von Schuckmann et al., 2020). In addition to the TOA energy fluxes, Earth’s energy budget al.o includes the internal flows of energy within the climate system, which characterize the climate state. The surface energy budget consists of the net solar and thermal radiation as well as the non-radiative components such as sensible, latent and ground heat fluxes (Figure 7.2, upper panel). It is a key driver of the global water cycle, atmosphere and ocean dynamics, as well as a variety of surface processes.

7.2.1 Present-day Energy Budget

Figure 7.2 (upper panel) shows a schematic representation of Earth’s energy budget for the early 21st century, including globally averaged estimates of the individual components (Wild et al., 2015). Clouds are important modulators of global energy fluxes. Thus, any perturbations in the cloud fields, such as forcing by aerosol–cloud interactions (Section 7.3) or through cloud feedbacks (Section 7.4) can have a strong influence on the energy distribution in the climate system. To illustrate the overall effects that clouds exert on energy fluxes, Figure 7.2 (lower panel) also shows the energy budget in the absence of clouds, with otherwise identical atmospheric and surface radiative properties. It has been derived by taking into account information contained in both in situ and satellite radiation measurements taken under cloud-free conditions (Wild et al., 2019). A comparison of the upper and lower panels in Figure 7.2 shows that without clouds, 47 W m–2 less solar radiation is reflected back to space globally (53 ± 2 W m–2 instead of 100 ± 2 W m–2), while 28 W m–2 more thermal radiation is emitted to space (267 ± 3 W m–2 instead of 239 ± 3 W m–2). As a result, there is a 20 W m–2 radiative imbalance at the TOA in the clear-sky energy budget (Figure 7.2, lower panel), suggesting that the Earth would warm substantially if there were no clouds.

The AR5 (Church et al., 2013; Hartmann et al., 2013; Myhre et al., 2013b) highlighted the progress that had been made in quantifying the TOA radiation budget following new satellite observations that became available in the early 21st century (Clouds and the Earth’s Radiant Energy System, CERES; Solar Radiation and Climate Experiment, SORCE). Progress in the quantification of changes in incoming solar radiation at the TOA is discussed in Chapter 2 (Section 2.2). Since AR5, the CERES Energy Balance EBAF Ed4.0 product was released, which includes algorithm improvements and consistent input datasets throughout the record (Loeb et al., 2018b). However, the overall precision of these fluxes (uncertainty in global mean TOA flux of 1.7% (1.7 W m–2) for reflected solar and 1.3% (3.0 W m–2) for outgoing thermal radiation at the 90% confidence level) is not sufficient to quantify the Earth’s energy imbalance in absolute terms. Therefore, the CERES EBAF reflected solar and emitted thermal TOA fluxes were adjusted, within the estimated uncertainties, to ensure that the net TOA flux for July 2005 to June 2015 was consistent with the estimated Earth’s energy imbalance for the same period based on ocean heat content (OHC) measurements and energy uptake estimates for the land, cryosphere and atmosphere (Section 7.2.2.2; Johnson et al., 2016; Riser et al., 2016). ESMs typically show good agreement with global mean TOA fluxes from CERES-EBAF. However, as some ESMs are known to calibrate their TOA fluxes to CERES or similar data (Hourdin et al., 2017), this is not necessarily an indication of model accuracy, especially as ESMs show significant discrepancies on regional scales, often related to their representation of clouds (Trenberth and Fasullo, 2010; Donohoe and Battisti, 2012; Hwang and Frierson, 2013; J.-L.F. Li et al., 2013; Dolinar et al., 2015; Wild et al., 2015).

Figure 7.2 | Schematic representation of the global mean energy budget of the Earth (upper panel), and its equivalent without considerations of cloud effects (lower panel). Numbers indicate best estimates for the magnitudes of the globally averaged energy balance components in W m–2 together with their uncertainty ranges in parentheses (5–95% confidence range), representing climate conditions at the beginning of the 21st century. Note that the cloud-free energy budget shown in the lower panel is not the one that Earth would achieve in equilibrium when no clouds could form. It rather represents the global mean fluxes as determined solely by removing the clouds but otherwise retaining the entire atmospheric structure. This enables the quantification of the effects of clouds on the Earth energy budget and corresponds to the way clear-sky fluxes are calculated in climate models. Thus, the cloud-free energy budget is not closed and therefore the sensible and latent heat fluxes are not quantified in the lower panel. Figure adapted from Wild et al. (2015, 2019).

The radiation components of the surface energy budget are associated with substantially larger uncertainties than at the TOA, since they are less directly measured by passive satellite sensors and require retrieval algorithms and ancillary data for their estimation (Raschke et al., 2016; Kato et al., 2018; Huang et al., 2019). Confidence in the quantification of the global mean surface radiation components has increased recently, as independent estimates now converge to within a few W m–2(Wild, 2017). Current best estimates for downward solar and thermal radiation at Earth’s surface are approximately 185 W m–2 and 342 W m–2, respectively (Figure 7.2). These estimates are based on complementary approaches that make use of satellite products from active and passive sensors (L’Ecuyer et al., 2015; Kato et al., 2018) and information from surface observations and Earth system models (ESMs; Wild et al., 2015). Inconsistencies in the quantification of the global mean energy and water budgets discussed in AR5 (Hartmann et al., 2013) have been reconciled within the (considerable) uncertainty ranges of their individual components (Wild et al., 2013, 2015; L’Ecuyer et al., 2015). However, on regional scales, the closure of the surface energy budgets remains a challenge with satellite-derived datasets (Loeb et al., 2014; L’Ecuyer et al., 2015; Kato et al., 2016). Nevertheless, attempts have been made to derive surface energy budgets over land and ocean (Wild et al., 2015), over the Arctic (Christensen et al., 2016b), and over individual continents and ocean basins (L’Ecuyer et al., 2015; Thomas et al., 2020). Since AR5, the quantification of the uncertainties in surface energy flux datasets has improved. Uncertainties in global monthly mean downward solar and thermal fluxes in the CERES-EBAF surface dataset are, respectively, 10 W m–2 and 8 W m–2(converted to 5–95% ranges; Kato et al., 2018). The uncertainty in the surface fluxes for polar regions is larger than in other regions (Kato et al., 2018) due to the limited number of surface sites and larger uncertainty in surface observations (Previdi et al., 2015). The uncertainties in ocean mean latent and sensible heat fluxes are approximately 11 W m–2 and 5 W m–2(converted to 5–95% ranges), respectively (L’Ecuyer et al., 2015). A recent review of the latent and sensible heat flux accuracies over the period 2000–2007 highlights significant differences between several gridded products over ocean, where root-mean-squared differences between the multi-product ensemble and data at more than 200 moorings reached up to 25 W m–2 for latent heat and 5 W m–2 for sensible heat (Bentamy et al., 2017). This uncertainty stems from the retrieval of flux-relevant meteorological variables, as well as from differences in the flux parametrizations (Yu, 2019). Estimating the uncertainty in sensible and latent heat fluxes over land is difficult because of the large temporal and spatial variability. The flux values over land computed with three global datasets vary by 10–20% (L’Ecuyer et al., 2015).

ESMs also show larger discrepancies in their surface energy fluxes than at the TOA due to weaker observational constraints, with a spread of typically 10–20 W m–2 in the global average, and an even greater spread at regional scales (J.-L.F. Li et al., 2013; Wild et al., 2013; Boeke and Taylor, 2016; Wild, 2017, 2020; C. Zhang et al., 2018). Differences in the land-averaged downward thermal and solar radiation in CMIP5 ESMs amount to more than 30 and 40 W m–2, respectively (Wild et al., 2015). However, in the global multi-model mean, the magnitudes of the energy budget components of the CMIP6 ESMs generally show better agreement with reference estimates than previous model generations (Wild, 2020).

In summary, since AR5, the magnitudes of the global mean energy budget components have been quantified more accurately, not only at the TOA, but also at the Earth’s surface, where independent estimates of the radiative components have converged (high confidence). Considerable uncertainties remain in regional surface energy budget estimates as well as their representation in climate models.

7.2.2 Changes in Earth’s Energy Budget

7.2.2.1 Changes in Earth’s Top-of-atmosphere Energy Budget

Since 2000, changes in top-of-atmosphere (TOA) energy fluxes can be tracked from space using CERES satellite observations (Figure 7.3). The variations in TOA energy fluxes reflect the influence of internal climate variability, particularly that of El Niño–Southern Oscillation (ENSO), in addition to radiative forcing of the climate system and climate feedbacks (Allan et al., 2014; Loeb et al., 2018b). For example, globally, the reduction in both outgoing thermal and reflected solar radiation during La Niña conditions in 2008/2009 led to an energy gain for the climate system, whereas enhanced outgoing thermal and reflected solar radiation caused an energy loss during the El Niños of 2002/2003 and 2009/2010 (Figure 7.3; Loeb et al., 2018b). An ensemble of CMIP6 models is able to track the variability in global mean TOA fluxes observed by CERES, when driven with prescribed sea surface temperatures (SSTs) and sea ice concentrations (Figure 7.3; Loeb et al., 2020). Under cloud-free conditions, the CERES record shows a near zero trend in outgoing thermal radiation (Loeb et al., 2018b), which – combined with an increasing surface upwelling thermal flux – implies an increasing clear-sky greenhouse effect (Raghuraman et al., 2019). Conversely, clear-sky solar reflected TOA radiation in the CERES record covering March 2000 to September 2017 shows a decrease due to reductions in aerosol optical depth in the Northern Hemisphere and sea ice fraction (Loeb et al., 2018a; Paulot et al., 2018).

Figure 7.3 | Anomalies in global mean all-sky top-of-atmosphere (TOA) fluxes from CERES-EBAF Ed4.0 (solid black lines) and various CMIP6 climate models (coloured lines) in terms of (a) reflected solar, (b) emitted thermal and (c) net TOA fluxes. The multi-model means are additionally depicted as solid red lines. Model fluxes stem from simulations driven with prescribed sea surface temperatures (SSTs) and all known anthropogenic and natural forcings. Shown are anomalies of 12-month running means. All flux anomalies are defined as positive downwards, consistent with the sign convention used throughout this chapter. The correlations between the multi-model means (solid red lines) and the CERES records (solid black lines) for 12-month running means are: 0.85 for the global mean reflected solar; 0.73 for outgoing thermal radiation; and 0.81 for net TOA radiation. Figure adapted from Loeb et al. (2020). Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).

An effort to reconstruct variations in net TOA fluxes back to 1985, based on a combination of satellite data, atmospheric reanalysis and high-resolution climate model simulations (Allan et al., 2014; Liu et al., 2020), exhibits strong interannual variability associated with the volcanic eruption of Mount Pinatubo in 1991 and the ENSO events before 2000. The same reconstruction suggests that Earth’s energy imbalance increased by several tenths of a W m–2 between the periods 1985–1999 and 2000–2016, in agreement with the assessment of changes in the global energy inventory (Section 7.2.2.2, and Box 7.2, Figure 1). Comparisons of year-to-year variations in Earth’s energy imbalance estimated from CERES and independent estimates based on ocean heat content change are significantly correlated with similar phase and magnitude (Johnson et al., 2016; Meyssignac et al., 2019), promoting confidence in both satellite and in situ-based estimates (Section 7.2.2.2).

In summary, variations in the energy exchange between Earth and space can be accurately tracked since the advent of improved observations since the year 2000 (high confidence), while reconstructions indicate that the Earth’s energy imbalance was larger in the 2000s than in the 1985–1999 period (high confidence).

7.2.2.2 Changes in the Global Energy Inventory

The global energy inventory quantifies the integrated energy gain of the climate system associated with global ocean heat uptake, warming of the atmosphere, warming of the land, and melting of ice. Due to energy conservation, the rate of accumulation of energy in the Earth system (Section 7.1) is equivalent to the Earth energy imbalance (ΔN in Box 7.1, Equation 7.1). On annual and longer time scales, changes in the global energy inventory are dominated by changes in global ocean heat content (OHC; Rhein et al., 2013; Palmer and McNeall, 2014; Johnson et al., 2016). Thus, observational estimates and climate model simulations of OHC change are critical to the understanding of both past and future climate change (Sections 2.3.3.1, 3.5.1.3, 4.5.2.1 and 9.2.2.1).

Since AR5, both modelling and observation-based studies have established Earth’s energy imbalance (characterized by OHC change) as a more robust metric of the rate of global climate change than GSAT on interannual-to-decadal time scales (Palmer and McNeall, 2014; von Schuckmann et al., 2016; Wijffels et al., 2016; Cheng et al., 2018; Allison et al., 2020). This is because GSAT is influenced by large unforced variations, for example linked to ENSO and Pacific Decadal Variability (Roberts et al., 2015; Yan et al., 2016; Cheng et al., 2018). Measuring OHC change more comprehensively over the full ocean depth results in a higher signal-to-noise ratio and a time series that increases steadily over time (Box 7.2, Figure 1; Allison et al., 2020). In addition, understanding of the potential effects of historical ocean sampling on estimated global ocean heating rates has improved (Durack et al., 2014; Good, 2017; Allison et al., 2019) and there are now more estimates of OHC change available that aim to mitigate the effect of limited observational sampling in the Southern Hemisphere (Lyman and Johnson, 2008; Cheng et al., 2017; Ishii et al., 2017).

The assessment of changes in the global energy inventory for the periods 1971–2018, 1993–2018 and 2006–2018 draws upon the latest observational time series and the assessments presented in other chapters of this report. The estimates of OHC change come directly from the assessment presented in (Chapter 2 Section 2.3.3.1). The assessment of land and atmospheric heating comes from von Schuckmann et al. (2020), based on the estimates of Cuesta-Valero et al. (2021) and Steiner et al. (2020), respectively. Heating of inland waters, including lakes, reservoirs and rivers, is estimated to account for <0.1% of the total energy change, and is therefore omitted from this assessment (Vanderkelen et al., 2020). The cryosphere contribution from the melting of grounded ice is based on the mass-loss assessments presented in Chapter 9, Section 9.4.1 (Greenland Ice Sheet), Section 9.4.2 (Antarctic Ice Sheet) and (Section 9.5.1 (glaciers). Following AR5, the estimate of heating associated with loss of Arctic sea ice is based on a reanalysis (Schweiger et al., 2011), following the methods described by Slater et al. (2021). Chapter 9 Section 9.3.2) finds no significant trend in Antarctic sea ice area over the observational record, so a zero contribution is assumed. Ice melt associated with the calving and thinning of floating ice shelves is based on the decadal rates presented in Slater et al. (2021). For all cryospheric components, mass loss is converted to heat input using a latent heat of fusion of 3.34 × 105J Kg–1°C–1 with the second-order contributions from variations associated with ice type and warming of ice from sub-freezing temperatures disregarded, as in AR5. The net change in energy, quantified in Zettajoules (1 ZJ = 1021Joules), is computed for each component as the difference between the first and last year of each period (Table 7.1). The uncertainties in the depth-interval contributions to OHC are summed to get the uncertainty in global OHC change. All other uncertainties are assumed to be independent and added in quadrature.

Table 7.1 | Contributions of the different components of the global energy inventory for the periods 1971–2018, 1993–2018 and 2006–2018 (Box 7.2 and Cross-Chapter Box 9.1). Energy changes are computed as the difference between annual mean values or year mid-points. The total heating rates correspond to Earth’s energy imbalance and are expressed per unit area of Earth’s surface.

Component

1971–2018

1993–2018

2006–2018

Energy Gain (ZJ)

%

Energy Gain (ZJ)

%

Energy Gain (ZJ)

%

Ocean

0–700 m

700–2000 m

>2000 m

396.0 [285.7 to 506.2]

241.6 [162.7 to 320.5]

123.3 [96.0 to 150.5]

31.0 [15.7 to 46.4]

91.0

55.6

28.3

7.1

263.0 [194.1 to 331.9]

151.5 [114.1 to 188.9]

82.8 [59.9 to 105.6]

28.7 [14.5 to 43.0]

91.0

52.4

28.6

10.0

138.8 [86.4 to 191.3]

75.4 [48.7 to 102.0]

49.7 [29.0 to 70.4]

13.8 [7.0 to 20.6]

91.1

49.5

32.6

9.0

Land

21.8 [18.6 to 25.0]

5.0

13.7 [12.4 to 14.9]

4.7

7.2 [6.6 to 7.8]

4.7

Cryosphere

11.5 [9.0 to 14.0]

2.7

8.8 [7.0 to 10.5]

3.0

4.7 [3.3 to 6.2]

3.1

Atmosphere

5.6 [4.6 to 6.7]

1.3

3.8 [3.2 to 4.3]

1.3

1.6 [1.2 to 2.1]

1.1

TOTAL

434.9 [324.5 to 545.3] ZJ

289.2 [220.3 to 358.1] ZJ

152.4 [100.0 to 204.9] ZJ

Heating Rate

0.57 [0.43 to 0.72] W m–2

0.72 [0.55 to 0.89] W m–2

0.79 [0.52 to 1.06] W m–2

For the period 1971–2010, AR5 (Rhein et al., 2013) found an increase in the global energy inventory of 274 [196 to 351] ZJ with a 93% contribution from total OHC change, approximately 3% for both ice melt and land heating, and approximately 1% for warming of the atmosphere. For the same period, this Report finds an upwards revision of OHC change for the upper (<700 m depth) and deep (>700 m depth) ocean of approximately 8% and 20%, respectively, compared to AR5 and a modest increase in the estimated uncertainties associated with the ensemble approach of Palmer et al. (2021). The other substantive change compared to AR5 is the updated assessment of land heating, with values approximately double those assessed previously, based on a more comprehensive analysis of the available observations (von Schuckmann et al., 2020; Cuesta-Valero et al., 2021). The result of these changes is an assessed energy gain of 329 [224 to 434] ZJ for the period 1971–2010, which is consistent with AR5 within the estimated uncertainties, despite the systematic increase.

The assessed changes in the global energy inventory (Box 7.2, Figure 1, and Table 7.1) yields an average value for Earth’s energy imbalance (N in Box 7.1, Equation 7.1) of 0.57 [0.43 to 0.72] W m–2 for the period 1971–2018, expressed relative to Earth’s surface area (high confidence). The estimates for the periods 1993–2018 and 2006–2018 yield substantially larger values of 0.72 [0.55 to 0.89] W m–2 and 0.79 [0.52 to 1.06] W m–2, respectively, consistent with the increased radiative forcing from GHGs (high confidence). For the period 1971–2006, the total energy gain was 282 [177 to 387] ZJ, with an equivalent Earth energy imbalance of 0.50 [0.32 to 0.69] W m–2. To put these numbers in context, the 2006–2018 average Earth system heating is equivalent to approximately 20 times the annual rate of global energy consumption in 2018. 1

Consistent with AR5 (Rhein et al., 2013), this Report finds that ocean warming dominates the changes in the global energy inventory (high confidence), accounting for 91% of the observed change for all periods considered (Table 7.1). The contributions from the other components across all periods are approximately 5% from land heating, 3% for cryosphere heating and 1% associated with warming of the atmosphere (high confidence). The assessed percentage contributions are similar to the recent study by von Schuckmann et al. (2020) and the total heating rates are consistent within the assessed uncertainties. Cross-validation of heating rates based on satellite and in situ observations (Section 7.2.2.1), and closure of the global sea level budget using consistent datasets (Cross-Chapter Box 9.1 and Table 9.5), strengthen scientific confidence in the assessed changes in the global energy inventory relative to AR5.

7.2.2.3 Changes in Earth’s Surface Energy Budget

The AR5 (Hartmann et al., 2013) reported pronounced changes in multi-decadal records of in situ observations of surface solar radiation, including a widespread decline between the 1950s and 1980s, known as ‘global dimming’, and a partial recovery thereafter, termed ‘brightening’ Section 12.4). These changes have interacted with closely related elements of climate change, such as global and regional warming rates (Z. Li et al., 2016; Wild, 2016; Du et al., 2017; Zhou et al., 2018a), glacier melt (Ohmura et al., 2007; Huss et al., 2009), the intensity of the global water cycle (Wild, 2012) and terrestrial carbon uptake (Mercado et al., 2009). These observed changes have also been used as emergent constraints to quantify aerosol effective radiative forcing (Section 7.3.3.3).

Since AR5, additional evidence for dimming and/or subsequent brightening up to several percent per decade, based on direct surface observations, has been documented in previously less-studied areas of the globe, such as Iran, Bahrain, Tenerife, Hawaii, the Taklaman Desert and the Tibetan Plateau (Elagib and Alvi, 2013; You et al., 2013; Garcia et al., 2014; Longman et al., 2014; Rahimzadeh et al., 2015). Strong decadal trends in surface solar radiation remain evident after careful data quality assessment and homogenization of long-term records (Sanchez-Lorenzo et al., 2013, 2015; Manara et al., 2015, 2016; Wang et al., 2015; Z. Li et al., 2016; Wang and Wild, 2016; Y. He et al., 2018; Yang et al., 2018). Since AR5, new studies on the potential effects of urbanization on solar radiation trends indicate that these effects are generally small, with the exception of some specific sites in Russia and China (Wang et al., 2014; Imamovic et al., 2016; Tanaka et al., 2016). Also, surface-based solar radiation observations have been shown to be representative over large spatial domains of up to several degrees latitude/longitude on monthly and longer time scales (Hakuba et al., 2014; Schwarz et al., 2018). Thus, there is high confidence that the observed dimming between the 1950s and 1980s and the subsequent brightening are robust and do not arise from measurement artefacts or localized phenomena.

As noted in AR5 (Hartmann et al., 2013) and supported by recent studies, the trends in surface solar radiation are less spatially coherent since the beginning of the 21st century, with evidence for continued brightening in parts of Europe and the USA, some stabilization in China and India, and dimming in other areas (Augustine and Dutton, 2013; Sanchez-Lorenzo et al., 2015; Manara et al., 2016; Soni et al., 2016; Wang and Wild, 2016; Jahani et al., 2018; Pfeifroth et al., 2018; Yang et al., 2018; Schwarz et al., 2020). The CERES-EBAF satellite-derived dataset of surface solar radiation (Kato et al., 2018) does not indicate a globally significant trend over the short period 2001–2012 (Zhang et al., 2015), whereas a statistically significant increase in surface solar radiation of +3.4 W m−2 per decade over the period 1996–2010 has been found in the Satellite Application Facility on Climate Monitoring (CM SAF) record of the geostationary satellite Meteosat, which views Europe, Africa and adjacent ocean (Posselt et al., 2014).

Since AR5, there is additional evidence that strong decadal changes in surface solar radiation have occurred under cloud-free conditions, as shown for long-term observational records in Europe, USA, China, India and Japan (Xu et al., 2011; Gan et al., 2014; Manara et al., 2016; Soni et al., 2016; Tanaka et al., 2016; Kazadzis et al., 2018; J. Li et al., 2018; Yang et al., 2019; Wild et al., 2021). This suggests that changes in the composition of the cloud-free atmosphere, primarily in aerosols, contributed to these variations, particularly since the second half of the 20th century (Wild, 2016). Water vapour and other radiatively active gases seem to have played a minor role (Wild, 2009; Mateos et al., 2013; Posselt et al., 2014; Yang et al., 2019). For Europe and East Asia, modelling studies also point to aerosols as an important factor for dimming and brightening by comparing simulations that include or exclude variations in anthropogenic aerosol and aerosol-precursor emissions (Golaz et al., 2013; Nabat et al., 2014; Persad et al., 2014; Folini and Wild, 2015; Turnock et al., 2015; Moseid et al., 2020). Moreover, decadal changes in surface solar radiation have often occurred in line with changes in anthropogenic aerosol emissions and associated aerosol optical depth (Streets et al., 2006; Wang and Yang, 2014; Storelvmo et al., 2016; Wild, 2016; Kinne, 2019). However, further evidence for the influence of changes in cloudiness on dimming and brightening is emphasized in some studies (Augustine and Dutton, 2013; Parding et al., 2014; Stanhill et al., 2014; Pfeifroth et al., 2018; Antuña-Marrero et al., 2019). Thus, the contribution of aerosol and clouds to dimming and brightening is still debated. The relative influence of cloud-mediated aerosol effects versus direct aerosol radiative effects on dimming and brightening in a specific region may depend on the prevailing pollution levels (Section 7.3.3; Wild, 2016).

ESMs and reanalyses often do not reproduce the full extent of observed dimming and brightening (Wild and Schmucki, 2011; Allen et al., 2013; Zhou et al., 2017a; Storelvmo et al., 2018; Moseid et al., 2020; Wohland et al., 2020), potentially pointing to inadequacies in the representation of aerosol mediated effects or related emissions data. The inclusion of assimilated aerosol optical depth inferred from satellite retrievals in the MERRA2 reanalysis (Buchard et al., 2017; Randles et al., 2017) helps to improve the accuracy of the simulated surface solar radiation changes in China (Feng and Wang, 2019). However, non-aerosol-related deficiencies in model representations of clouds and circulation, and/or an underestimation of natural variability, could further contribute to the lack of dimming and brightening in ESMs (Wild, 2016; Storelvmo et al., 2018).

The AR5 reported evidence for an increase in surface downward thermal radiation based on different studies covering 1964 to 2008, in line with what would be expected from an increased radiative forcing from GHGs and the warming and moistening of the atmosphere. Updates of the longest observational records from the Baseline Surface Radiation Network continue to show an increase at the majority of sites, in line with an overall increase predicted by ESMs of the order of 2 W m–2 per decade (Wild, 2016). Upward longwave radiation at the surface is rarely measured but is expected to have increased over the same period due to rising surface temperatures.

Turbulent fluxes of latent and sensible heat are also an important part of the surface energy budget (Figure 7.2). Large uncertainties in measurements of surface turbulent fluxes continue to prevent the determination of their decadal changes. Nevertheless, over the ocean, reanalysis-based estimates of linear trends from 1948–2008 indicate high spatial variability and seasonality. Increases in magnitudes of 4 to 7 W m–2 per decade for latent heat and 2 to 3 W m–2 per decade for sensible heat in the western boundary current regions are mostly balanced by decreasing trends in other regions (Gulev and Belyaev, 2012). Over land, the terrestrial latent heat flux is estimated to have increased in magnitude by 0.09 W m–2 per decade from 1989–1997, and subsequently decreased by 0.13 W m–2 per decade from 1998–2005 due to soil-moisture limitation mainly in the Southern Hemisphere (derived from Mueller et al., 2013). These trends are small in comparison to the uncertainty associated with satellite-derived and in situ observations, as well as from land-surface models forced by observations and atmospheric reanalyses. Ongoing advances in remote sensing of evapotranspiration from space (Mallick et al., 2016; Fisher et al., 2017; McCabe et al., 2017a, b), as well as terrestrial water storage (Rodell et al., 2018) may contribute to future constraints on changes in latent heat flux.

In summary, since AR5, multi-decadal decreasing and increasing trends in surface solar radiation of up to several percent per decade have been detected at many more locations, even in remote areas. There is high confidence that these trends are widespread, and not localized phenomena or measurement artefacts. The origin of these trends is not fully understood, although there is evidence that anthropogenic aerosols have made a substantial contribution (medium confidence). There is medium confidence that downward and upward thermal radiation has increased since the 1970s, while there remains low confidence in the trends in surface sensible and latent heat.

Box 7.2 | The Global Energy Budget

This box assesses the present knowledge of the global energy budget for the period 1971–2018, that is, the balance between radiative forcing, the climate system radiative response and observations of the changes in the global energy inventory (Box 7.2, Figure 1a,d).

The net effective radiative forcing (ERF) of the Earth system since 1971 has been positive (Section 7.3 and Box 7.2, Figure 1b,e), mainly as a result of increases in atmospheric greenhouse gas concentrations (Sections 2.2.8 and 7.3.2). The ERF of these positive forcing agents have been partly offset by that of negative forcing agents, primarily due to anthropogenic aerosols (Section 7.3.3), which dominate the overall uncertainty. The net energy inflow to the Earth system from ERF for the period 1971–2018 is estimated to be 937 ZJ (1 ZJ = 1021J) with a likely range of 644 to 1259 ZJ (Box 7.2, Figure 1b).

Box 7.2

The ERF-induced heating of the climate system results in increased thermal radiation to space via the Planck response, but the picture is complicated by a variety of climate feedbacks (Section 7.4.2 and Box 7.1) that also influence the climate system radiative response (Box 7.2, Figure 1c). The total radiative response is estimated by multiplying the assessed net feedback parameter, α , from process-based evidence (Section 7.4.2 and Table 7.10) with the observed GSAT change for the period (Cross Chapter Box 2.3) and time-integrating (Box 7.2, Figure 1c). The net energy outflow from the Earth system associated with the integrated radiative response for the period 1971–2018 is estimated to be 621 ZJ with a likely range of 419 to 823 ZJ. Assuming a pattern effect (Section 7.4.4) on α of –0.5 W m–2°C–1 would lead to a systematically larger energy outflow by about 250 ZJ.

Box 7.2, Figure 1 | Estimates of the net cumulative energy change (ZJ = 1021Joules) for the period 1971–2018 associated with: (a) observations of changes in the global energy inventory; (b) integrated radiative forcing; and (c) integrated radiative response. Black dotted lines indicate the central estimate with likely and very likely ranges as indicated in the legend. The grey dotted lines indicate the energy change associated with an estimated pre-industrial Earth energy imbalance of 0.2 W m–2(a), and an illustration of an assumed pattern effect of –0.5 W m–2°C–1(c). Background grey lines indicate equivalent heating rates in W m–2 per unit area of Earth’s surface. Panels (d) and (e) show the breakdown of components, as indicated in the legend, for the global energy inventory and integrated radiative forcing, respectively. Panel (f) shows the global energy budget assessed for the period 1971–2018, that is, the consistency between the change in the global energy inventory relative to pre-industrial and the implied energy change from integrated radiative forcing plus integrated radiative response under a number of different assumptions, as indicated in the legend, including assumptions of correlated and uncorrelated uncertainties in forcing plus response. Shading represents the very likely range for observed energy change relative to pre-industrial levels and likely range for all other quantities. Forcing and response time series are expressed relative to a baseline period of 1850–1900. Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).

Combining the likely range of integrated radiative forcing (Box 7.2, Figure 1b) with the central estimate of integrated radiative response (Box 7.2, Figure 1c) gives a central estimate and likely range of 340 [47 to 662] ZJ (Box 7.2, Figure 1f). Combining the likely range of integrated radiative response with the central estimate of integrated radiative forcing gives a likely range of 340 [147 to 527] ZJ (Box 7.2, Figure 1f). Both calculations yield an implied energy gain in the climate system that is consistent with an independent observation-based assessment of the increase in the global energy inventory expressed relative to the estimated 1850–1900 Earth energy imbalance (Section 7.5.2 and Box 7.2, Figure 1a) with a central estimate and very likely range of 284 [96 to 471] ZJ (high confidence) (Box 7.2, Figure 1d; Table 7.1). Estimating the total uncertainty associated with radiative forcing and radiative response remains a scientific challenge and depends on the degree of correlation between the two (Box 7.2, Figure 1f). However, the central estimate of observed energy change falls well with the estimated likely range, assuming either correlated or uncorrelated uncertainties. Furthermore, the energy budget assessment would accommodate a substantial pattern effect (Section 7.4.4.3) during 1971–2018 associated with systematically larger values of radiative response (Box 7.2, Figure 1c), and potentially improved closure of the global energy budget. For the period 1970–2011, AR5 reported that the global energy budget was closed within uncertainties (high confidence) and consistent with the likely range of assessed climate sensitivity (Church et al., 2013). This Report provides a more robust quantitative assessment based on additional evidence and improved scientific understanding.

In addition to new and extended observations (Section 7.2.2), confidence in the observed accumulation of energy in the Earth system is strengthened by cross-validation of heating rates based on satellite and in situ observations (Section 7.2.2.1) and closure of the global sea level budget using consistent datasets (Cross-Chapter Box 9.1 and Table 9.5). Overall, there is high confidence that the global energy budget is closed for 1971–2018 with improved consistency compared to AR5.

7.3 Effective Radiative Forcing

Effective radiative forcing (ERF) quantifies the energy gained or lost by the Earth system following an imposed perturbation (for instance in GHGs, aerosols or solar irradiance). As such it is a fundamental driver of changes in the Earth’s TOA energy budget. ERF is determined by the change in the net downward radiative flux at the TOA (Box 7.1) after the system has adjusted to the perturbation but excluding the radiative response to changes in surface temperature. This section outlines the methodology for ERF calculations (Section 7.3.1) and then assesses the ERF due to greenhouse gases (Section 7.3.2), aerosols (Section 7.3.3) and other natural and anthropogenic forcing agents (Section 7.3.4). These are brought together in (Section 7.3.5 for an overall assessment of the present-day ERF and its evolution over the historical time period from 1750 to 2019. The same section also evaluates the surface temperature response to individual ERFs.

7.3.1 Methodologies and Representation in Models: Overview of Adjustments

As introduced in Box 7.1, AR5 (Boucher et al., 2013; Myhre et al., 2013b) recommended ERF as a more useful measure of the climate effects of a physical driver than the stratospheric-temperature-adjusted radiative forcing (SARF) adopted in earlier assessments. The AR5 assessed that the ratios of surface temperature change to forcing resulting from perturbations of different forcing agents were more similar between species using ERF than SARF. ERF extended the SARF concept to account for not only adjustments to stratospheric temperatures, but also responses in the troposphere and effects on clouds and atmospheric circulation, referred to as ‘adjustments’. For more details see Box 7.1. Since circulation can be affected, these responses are not confined to the locality of the initial perturbation (unlike the traditional stratospheric-temperature adjustment).

This chapter defines ‘adjustments’ as those changes caused by the forcing agent that are independent of changes in surface temperature, rather than defining a specific time scale. The AR5 used the term ‘rapid adjustment’, but in this assessment the definition is based on the independence from surface temperature rather than the rapidity. The definition of ERF in Box 7.1 aims to create a clean separation between forcing (energy budget changes that are not mediated by surface temperature) and feedbacks (energy budget changes that are mediated by surface temperature). This means that changes in land or ocean surface temperature patterns (for instance as identified by Rugenstein et al., 2016b) are not included as adjustments. As in previous assessments (Forster et al., 2007; Myhre et al., 2013b) ERFs can be attributed simply to changes in the forcing agent itself or attributed to components of emitted gases (Figure 6.12). Because ERFs can include chemical and biospheric responses to emitted gases, they can be attributed to precursor gases, even if those gases do not have a direct radiative effect themselves. Similar chemical and biospheric responses to forcing agents can also be included in the ERF in addition to their direct effects.

Instantaneous radiative forcing (IRF) is defined here as the change in the net TOA radiative flux following a perturbation, excluding any adjustments. SARF is defined here as the change in the net radiative flux at TOA following a perturbation including the response to stratospheric temperature adjustments. These differ from AR5 where these quantities were defined at the tropopause (Myhre et al., 2013b). The net IRF values will be different using the TOA definition. The net SARF values will be the same as with the tropopause definition, but will have a different partitioning between the longwave and shortwave. Defining all quantities at the TOA enables consistency in breaking down the ERF into its component parts.

The assessment of ERFs in AR5 was preliminary because ERFs were only available for a few forcing agents, so for many forcing agents the Report made the assumption that ERF and SARF were equivalent. This section discusses the body of work published since AR5. This work has computed ERFs across many more forcing agents and models; closely examined the methods of computation; quantified the processes involved in causing adjustments; and examined how well ERFs predict the ultimate temperature response. This work is assessed to have led to a much-improved understanding and increased confidence in the quantification of radiative forcing across the Report. These same techniques allow for an evaluation of radiative forcing within Earth system models (ESMs) as a key test of their ability to represent both historical and future temperature changes (Sections 3.3.1 and 4.3.4).

The ERF for a particular forcing agent is the sum of the IRF and the contribution from the adjustments, so in principle this could be constructed bottom-up by calculating the IRF and adding in the adjustment contributions one-by-one or together. However, there is no simple way to derive the global tropospheric adjustment terms or adjustments related to circulation changes without using a comprehensive climate model (e.g., CMIP5 or CMIP6). There have been two main modelling approaches used to approximate the ERF definition in Box 7.1. The first approach is to use the assumed linearity (Box 7.1, Equation 7.1) to regress the net change in the TOA radiation budget (ΔN) against change in global mean surface temperature (ΔT) following a step change in the forcing agent (Box 7.1, Figure 1; Gregory et al., 2004). The ERF (ΔF) is then derived from ΔN when ΔT= 0. Regression-based estimates of ERF depend on the temporal resolution of the data used (Modak et al., 2016, 2018). For the first few months of a simulation both surface temperature change and stratospheric-temperature adjustment occur at the same time, leading to misattribution of the stratospheric-temperature adjustment to the surface temperature feedback. Patterns of sea surface temperature (SST) change also affect estimates of the forcing obtained by regression methods (Andrews et al., 2015). At multi-decadal time scales the curvature of the relationship between net TOA radiation and surface temperature can also lead to biases in the ERF estimated from the regression method (Section 7.4; Armour et al., 2013; Andrews et al., 2015; Knutti et al., 2017). The second modelling approach to estimate ERF is to set the ΔT term in Box 7.1 (Box 7.1, Equation 7.1) to zero. It is technically difficult to constrain land surface temperatures in ESMs (Shine et al., 2003; Ackerley and Dommenget, 2016; Andrews et al., 2021), so most studies reduce the ΔT term by prescribing the SSTs and sea ice concentrations in a pair of ‘fixed-SST’ (fSST) simulations with and without the change in forcing agent (Hansen et al., 2005b). An approximation to ERF (ΔFfsst ) is then given by the difference in ΔNfsst 4 between the simulations. The fSST method has less noise due to internal variability than the regression method. Nevertheless a 30-year fSST integration or 10 × 20-year regression ensemble needs to be conducted in order to reduce the 5–95% confidence range to 0.1 W m–2(Forster et al., 2016).Neither the regression or fSST methods are practical for quantifying the ERF of agents with forcing magnitudes of the order of 0.1 W m–2 or smaller. The internal variability in the fSST method can be further constrained by nudging winds towards a prescribed climatology (Kooperman et al., 2012). This allows the determination of the ERF of forcing agents with smaller magnitudes but excludes adjustments associated with circulation responses (Schmidt et al., 2018). There are insufficient studies to assess whether these circulation adjustments are significant.

Since the near-surface temperature change over land, ΔTland, is not constrained in the fSST method, this response needs to be removed for consistency with the (Section 7.1 definition. These changes in the near-surface temperature will also induce further responses in the tropospheric temperature and water vapour that should also be removed to conform with the physical definition of ERF. The radiative response to ΔTland can be estimated through radiative transfer modelling in which a kernel, k, representing the change in net TOA radiative flux per unit of change in near-surface temperature change over land (or an approximation using land surface temperature), is precomputed (Smith et al., 2018b, 2020b; Richardson et al., 2019; Tang et al., 2019). Thus ERF ≈ ΔFfsst kΔTland. Since k is negative this means that ΔFfsst underestimates the ERF. For 2×CO2 , this underestimate is around 0.2 W m–2(Smith et al., 2018b, 2020b). There have been estimates of the corrections due to tropospheric temperature and water vapour (Tang et al., 2019; Smith et al., 2020b) showing additional radiative responses of comparable magnitude to those directly from ΔTland. An alternative to computing the response terms directly is to use the feedback parameter, α (Hansen et al., 2005b; Sherwood et al., 2015; Tang et al., 2019). This gives approximately double the correction compared to the kernel approach (Tang et al., 2019). The response to land surface temperature change varies with location and even for GSAT change k is not expected to be the same as α Section 7.4). One study where land surface temperatures are constrained in a model (Andrews et al., 2021) finds this constraint adds +1.0 W m–2 to ΔFfsst for 4×CO2 , thus confirming the need for a correction in calculations where this constraint is not applied. For this assessment the correction is conservatively based only on the direct radiative response kernel to ΔTland as this has a strong theoretical basis to support it. While there is currently insufficient corroborating evidence to recommend including tropospheric temperature and water-vapour corrections in this assessment, it is noted that the science is progressing rapidly on this topic.

TOA radiative flux changes due to the individual adjustments can be calculated by perturbing the meteorological fields in a climate model’s radiative transfer scheme (partial radiative perturbation approach) (Colman, 2015; Mülmenstädt et al., 2019) or by using precomputed radiative kernels of sensitivities of the TOA radiation fluxes to changes in these fields (as done for near-surface temperature change above; Vial et al., 2013; Zelinka et al., 2014; Zhang and Huang, 2014; Smith et al., 2018b, 2020b). The radiative kernel approach is easier to implement through post-processing of output from multiple ESMs, whereas it is recognized that the partial radiation perturbation approach gives a more accurate estimate of the adjustments within the setup of a single model and its own radiative transfer code. There is little difference between using a radiative kernel from the same or a different model when calculating the adjustment terms, except for stratospheric temperature adjustments where it is important to have sufficient vertical resolution in the stratosphere in the model used to derive the kernel (Smith et al., 2018b, 2020a).

For comparison with offline radiative transfer calculations the SARFs can be approximated by removing the adjustment terms (apart from stratospheric temperature) from the ERFs using radiative kernels to quantify the adjustment for each meteorological variable. Kernel analysis by Chung and Soden (2015) suggested a large spread in CO2 SARF across climate models, but their analysis was based on regressing variables in a coupled-ocean experiment rather than using a fSST approach which leads to a large spread due to natural variability (Forster et al., 2016). Adjustments computed from radiative kernels are shown for seven different climate drivers (using a fSST approach) in Figure 7.4. Table 7.2 shows the estimates of SARF, ΔFfsst and ERF (corrected for land surface temperature change) for 2×CO2 from the nine climate models analysed in Smith et al. (2018b). The SARF shows a smaller spread over previous studies (Pincus et al., 2016; Soden et al., 2018) and most estimates are within 10% of the multi-model mean and the assessment of 2×CO2 SARF in (Section 7.3.2 (3.75 W m–2). It is not possible from these studies to determine how much of this reduction in spread is due to convergence in the model radiation schemes or the meteorological conditions of the model base states; nevertheless the level of agreement in this and earlier intercomparisons gives medium confidence in the ability of ESMs to represent radiative forcing from CO2. The 4×CO2 CMIP6 fSST experiments (Smith et al., 2020b) in Table 7.2 include ESMs with varying levels of complexity in aerosols and reactive gas chemistry. The CMIP6 experimental setup allows for further climate effects of CO2 (including on aerosols and ozone) depending on model complexity. The chemical effects are adjustments to CO2 but are not separable from the SARF in the diagnosis in Table 7.2. In these particular models, this leads to higher SARF than when only CO2 varies, however there are insufficient studies to make a formal assessment of composition adjustments to CO2.

Figure 7.4 | Radiative adjustments at top of atmosphere for seven different climate drivers as a proportion of forcing. Tropospheric temperature (orange), stratospheric temperature (yellow), water vapour (blue), surface albedo (green), clouds (grey) and the total adjustment (black) is shown. For the greenhouse gases (carbon dioxide, methane, nitrous oxide and CFC-12) the adjustments are expressed as a percentage of stratospheric-temperature-adjusted radiative forcing (SARF), whereas for aerosol, solar and volcanic forcing they are expressed as a percentage of instantaneous radiative forcing (IRF). Land surface temperature response (outline red bar) is shown, but included in the definition of forcing. Data from Smith et al. (2018b) for carbon dioxide and methane; Smith et al. (2018b) and Gray et al. (2009) for solar; Hodnebrog et al. (2020b) for nitrous oxide and CFC-12; Smith et al. (2020b) for aerosol, and Marshall et al. (2020) for volcanic. Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).
Table 7.2 | SARF, ΔFfsst , and ERF diagnosed from Earth system models for fixed-SST (fSST) CO2 experiments. 2×CO2 data taken from fixed atmospheric composition experiments (Smith et al., 2018b). 4×CO2 data taken from CMIP6 experiments with interactive aerosols (and interactive gas phase chemistry in some; Smith et al., 2020b). The radiative forcings from the 4×CO2 experiments are scaled by 0.476 for comparison with 2×CO2 (Meinshausen et al., 2020). SARF is approximated by removing the (non-stratospheric-temperature) adjustment terms from the ERF. In Smith et al. (2018b), separation of temperature adjustments into tropospheric and stratospheric contributions is approximate based on a fixed tropopause of 100 hPa at the equator, varying linearly in latitude to 300 hPa at the poles. In Smith et al. (2020b), this separation is based on the model-diagnosed tropopause. ERF is approximated by removing the response to land surface temperature change fromΔFfsst . The confidence range is based on the inter-model standard deviation.

2×CO2 Experiments

(Smith et al., 2018b)

Stratospheric- temperature-adjusted Radiative Forcing (SARF, W m–2)

ΔFfsst (W m–2)

Effective Radiative Forcing (ERF, W m–2)

HadGEM2-ES

3.45

3.37

3.58

NorESM1

3.67

3.50

3.70

GISS-E2-R

3.98

4.06

4.27

CanESM2

3.68

3.57

3.77

MIROC-SPRINTARS

3.89

3.62

3.82

NCAR-CESM1-CAM5

3.89

4.08

4.39

HadGEM3

3.48

3.64

3.90

IPSL-CM5A

3.50

3.39

3.61

MPI-ESM

4.27

4.14

4.38

NCAR-CESM1-CAM4

3.50

3.62

3.86

Multi-model mean and 5–95% confidence range

3.73 ± 0.44

3.70 ± 0.44

3.93 ± 0.48

0.476 × 4×CO2 Experiments

(Smith et al., 2020b)

Stratospheric- temperature-adjusted Radiative Forcing (SARF, W m–2)

ΔFfsst (W m–2)

Effective Radiative Forcing (ERF, W m–2)

ACCESS-CM2

3.56

3.78

3.98

CanESM5

3.67

3.62

3.82

CESM2

3.56

4.24

4.48

CNRM-CM6-1

3.99

3.81

4.01

CNRM-ESM2-1

3.99

3.77

3.94

EC-Earth3

3.85

4.04

GFDL-CM4

3.65

3.92

4.10

GFDL-ESM4

3.27

3.68

3.85

GISS-E2-1-G

3.78

3.50

3.69

HadGEM3-GC31-LL

3.61

3.85

4.07

IPSL-CM6A-LR

3.84

3.81

4.05

MIROC6

3.63

3.48

3.69

MPI-ESM1-2-LR

3.74

3.97

4.20

MRI-ESM2-0

3.76

3.64

3.80

NorESM2-LM

3.58

3.88

4.10

NorESM2-MM

3.62

3.99

4.22

UKESM1-0-LL

3.49

3.78

4.01

Multi-model mean and 5–95% confidence range

3.67 ± 0.29

3.80 ± 0.30

4.00 ± 0.32

ERFs have been found to yield more consistent values of GSAT change per unit forcing than SARF, that is, α shows less variation across different forcing agents (Rotstayn and Penner, 2001; Shine et al., 2003; Hansen et al., 2005b; Marvel et al., 2016; Richardson et al., 2019). Having a consistent relationship between forcing and response is advantageous when making climate projections using simple models (Cross-Chapter Box 7.1) or emissions metrics (Section 7.6). The definition of ERF used in this assessment, which excludes the radiative response to land surface temperature changes, brings The α values into closer agreement than when SARF is used (Richardson et al., 2019), although for individual models there are still variations, particularly for more geographically localized forcing agents. However, even for ERF, studies find that α is not identical across all forcing agents (Shindell, 2014; Shindell et al., 2015; Modak et al., 2018; Modak and Bala, 2019; Richardson et al., 2019). Section 7.4.4 discusses the effect of different SST response patterns on α . Analysis of the climate feedbacks (Kang and Xie, 2014; Gregory et al., 2016, 2020; Marvel et al., 2016; Duan et al., 2018; Persad and Caldeira, 2018; Stuecker et al., 2018; Krishnamohan et al., 2019) suggests a weaker feedback (i.e., less-negative α ) and hence larger sensitivity for forcing of the higher latitudes (particularly the Northern Hemisphere). Nonetheless, as none of these variations are robust across models, the ratio of 1/ α from non-CO2 forcing agents (with approximately global distributions) to that from doubling CO2 is within 10% of unity.

In summary, this Report adopts an estimate of ERF based on the change in TOA radiative fluxes in the absence of GSAT changes. This allows for a theoretically cleaner separation between forcing and feedbacks in terms of factors respectively unrelated and related to GSAT change (Box 7.1). ERF can be computed from prescribed SST and sea ice experiments after removing the TOA energy budget change associated with the land surface temperature response. In this assessment this is removed using a kernel accounting only for the direct radiative effect of the land surface temperature response. To compare these results with sophisticated high spectral resolution radiative transfer models the individual tropospheric adjustment terms can be removed to leave the SARF. SARFs for 2×CO2 calculated by ESMs from this method agree within 10% with the more sophisticated models. The new studies highlighted above suggest that physical feedback parameters computed within this framework have less variation across forcing agents. There is high confidence that an α based on ERF as defined here varies by less (less than variation 10% across a range of forcing agents with global distributions), than α based on SARF. For geographically localized forcing agents there are fewer studies and less agreement between them, resulting in low confidence that ERF is a suitable estimator of the resulting global mean near-surface temperature response .

7.3.2 Greenhouse Gases

High spectral resolution radiative transfer models provide the most accurate calculations of radiative perturbations due to greenhouse gases (GHGs), with errors in the instantaneous radiative forcing (IRF) of less than 1% (Mlynczak et al., 2016; Pincus et al., 2020). They can calculate IRFs with no adjustments, or SARFs by accounting for the adjustment of stratospheric temperatures using a fixed dynamical heating. It is not possible with offline radiation models to account for other adjustments. The high-resolution model calculations of SARF for carbon dioxide, methane and nitrous oxide have been updated since AR5, which were based on Myhre et al. (1998). The new calculations include the shortwave forcing from methane and updates to the water vapour continuum (increasing the total SARF of methane by 25%) and account for the absorption band overlaps between carbon dioxide and nitrous oxide (Etminan et al., 2016). The associated simplified expressions, from a re-fitting of the Etminan et al. (2016) results by Meinshausen et al. (2020), are given in Supplementary Material, Table 7.SM.1. The shortwave contribution to the IRF of methane has been confirmed independently (Collins et al., 2018). Since they incorporate known missing effects we assess the new calculations as being a more appropriate representation than Myhre et al. (1998).

As described in (Section 7.3.1, ERFs can be estimated using ESMs, however the radiation schemes in climate models are approximations to high spectral resolution radiative transfer models with variations and biases in results between the schemes (Pincus et al., 2015). Hence ESMs alone are not sufficient to establish ERF best estimates for the well-mixed GHGs (WMGHGs). This assessment therefore estimates ERFs from a combined approach that uses the SARF from radiative transfer models and adds the tropospheric adjustments derived from ESMs.

In AR5, the main information used to assess components of ERFs beyond SARF was from Vial et al. (2013) who found a near-zero non-stratospheric adjustment (without correcting for near-surface temperature changes over land) in 4×CO2 CMIP5 model experiments, with an uncertainty of ±10% of the total CO2 ERF. No calculations were available for other WMGHGs, so ERF was therefore assessed to be approximately equal to SARF (within 10%) for all WMGHGs.

The effect of WMGHGs in ESMs can extend beyond their direct radiative effects to include effects on ozone and aerosol chemistry and natural emissions of ozone and aerosol precursors, and in the case of CO2 to vegetation cover through physiological effects. In some cases these can have significant effects on the overall radiative budget changes from perturbing WMGHGs within ESMs (Myhre et al., 2013b; Zarakas et al., 2020; O’Connor et al., 2021; Thornhill et al., 2021a). These composition adjustments are further discussed in (Chapter 6 (Section 6.4.2).

7.3.2.1 Carbon Dioxide (CO2)

The SARF for carbon dioxide (CO2) has been slightly revised due to updates to spectroscopic data and inclusion of the absorption band overlaps between N2O and CO2 (Etminan et al., 2016). The formulae fitting to the Etminan et al. (2016) results in Meinshausen et al. (2020) are used. This increases the SARF due to doubling CO2 slightly from 3.71 W m–2 in AR5 to 3.75 W m–2. Tropospheric responses to CO2 in fSST experiments have been found to lead to an approximate balance in their radiative effects between an increased radiative forcing due to water vapour, cloud and surface-albedo adjustments and a decrease due to increased tropospheric temperature and land surface temperature response (Table 7.3; Vial et al., 2013; Zhang and Huang, 2014; Smith et al., 2018b, 2020b). The ΔFfsst includes any effects represented within the ESMs on tropospheric adjustments due to changes in evapotranspiration or leaf area (mainly affecting surface and boundary-layer temperature, low-cloud amount, and albedo) from the CO2 -physiological effects (Doutriaux-Boucher et al., 2009; Cao et al., 2010; T.B. Richardson et al., 2018). The effect on surface temperature (negative longwave response) is consistent with the expected physiological responses and needs to be removed for consistency with the ERF definition. The split between surface and tropospheric temperature responses was not reported in Vial et al. (2013) or Zhang and Huang (2014) but the total of surface and tropospheric temperature response agrees with Smith et al. (2018b, 2020b), givingmedium confidence in this decomposition. Doutriaux-Boucher et al. (2009) and Andrews et al. (2021) (using the same land surface model) find a 13% and 10% increase respectively in ERF due to the physiological responses to CO2. The physiological adjustments are therefore assessed to make a substantial contribution to the overall tropospheric adjustment for CO2 (high confidence), but there is insufficient evidence to provide a quantification of the split between physiological and thermodynamic adjustments. These forcing adjustments due to the effects of CO2 on plant physiology differ from the biogeophysical feedbacks due to the effects of temperature changes on vegetation discussed in (Section 7.4.2.5. The adjustment is assumed to scale with the SARF in the absence of evidence for non-linearity. The tropospheric adjustment is assessed from Table 7.3 to be +5% of the SARF with an uncertainty of 5%, which is added to the Meinshausen et al. (2020) formula for SARF. Due to the agreement between the studies and the understanding of the physical mechanisms there is medium confidence in the mechanisms underpinning the tropospheric adjustment, but low confidence in its magnitude .

The ERF from doubling CO2 (2×CO2) from the 1750 level (278 ppm; Section 2.2.3.3) is assessed to be 3.93 ± 0.47 W m–2(high confidence). Its assessed components are given in Table 7.4. The combined spectroscopic and radiative transfer modelling uncertainties give an uncertainty in the CO2 SARF of around 10% or less (Etminan et al., 2016; Mlynczak et al., 2016). The overall uncertainty in CO2 ERF is assessed as ±12%, as the more uncertain adjustments only account for a small fraction of the ERF (Table 7.3). The 2×CO2 ERF estimate is 0.2 W m–2 larger than using the AR5 formula (Myhre et al., 2013b) due to the combined effects of tropospheric adjustments which were assumed to be zero in AR5. CO2 concentrations have increased from 278 ppm in 1750 to 410 ppm in 2019 Section 2.2.3.3). The historical ERF estimate from CO2 is revised upwards from the AR5 value of 1.82 ± 0.38 W m–2(1750–2011) to 2.16 ± 0.26 W m–2(1750–2019) in this assessment, from a combination of the revisions described above (0.06 W m–2) and the 19 ppm rise in atmospheric concentrations between 2011 and 2019 (0.27 W m–2). The ESM estimates of 2×CO2 ERF (Table 7.2) lie within ±12% of the assessed value (apart from CESM2). The definition of ERF can also include further physiological effects – for instance on dust, natural fires and biogenic emissions from the land and ocean – but these are not typically included in the modelling setup for 2×CO2 ERF.

Table 7.3 | Adjustments to the top-of-atmosphere (TOA) carbon dioxide forcing due to changes in stratospheric temperature, surface and tropospheric temperatures, water vapour, clouds, and surface albedo, as a fraction of the stratospheric-temperature-adjusted radiative forcing (SARF). Effective radiative forcing (ERF) is defined in this Report as excluding the surface temperature response.

Percentage of SARF (source study)

Surface Temperature

Tropospheric Temperature

Stratospheric

Temperature

Surface Albedo

Water Vapour

Clouds

Troposphere

(Including Surface)

Troposphere

(Excluding Surface)

Vial et al. (2013)

–20% combined

N/A

2%

6%

11%

–1%

N/A

Zhang and Huang (2014)

–23% combined

26%

N/A

6%

16%

–1%

N/A

Smith et al. (2018b)

–6%

–16%

30%

3%

6%

12%

–1%

+5%

Smith et al. (2020b)

–6%

–15%

35%

3%

6%

15%

+3%

+9%

Table 7. 4 | Assessed effective radiative forcing (ERF), stratospheric-temperature-adjusted radiative forcing (SARF) and tropospheric adjustments to 2×CO2 change since pre-industrial times compared to the AR5 assessed range (Myhre et al., 2013b). Adjustments are due to changes in tropospheric temperatures, water vapour, clouds and surface albedo and land cover and are taken from Smith et al. (2018b) and assessed as a percentage of SARF (Table 7.3). Uncertainties are based on multi-model spread in Smith et al. (2018b). Note some of the uncertainties are anticorrelated, which means that they do not sum linearly.

2×CO2 Forcing

AR5

SARF/ERF (W m–2)

SARF

(W m–2)

Tropospheric Temperature Adjustment

(W m–2)

Water Vapour Adjustment

(W m–2)

Cloud Adjustment (W m–2)

Surface Albedo and Land-cover Adjustment (W m–2)

Total Tropospheric Adjustment (W m–2)

ERF

(W m–2)

2×CO2 ERF components

3.71

3.75

–0.60

0.22

0.45

0.11

0.18

3.93

5–95% uncertainty ranges as percentage of ERF

10% (SARF)

20% (ERF)

<10%

±6%

±4%

±7%

±2%

±7%

±12%

7.3.2.2 Methane (CH4)

The SARF for methane (CH4) has been substantially increased due to updates to spectroscopic data and inclusion of shortwave absorption (Etminan et al., 2016). Adjustments have been calculated in nine climate models by Smith et al. (2018b). Since CH4 is found to absorb in the shortwave near infrared, only adjustments from those models including this absorption are taken into account. For these models the adjustments act to reduce the ERF because the shortwave absorption leads to tropospheric heating and reductions in upper tropospheric cloud amounts. The adjustment is –14% ± 15%, which counteracts much of the increase in SARF identified by Etminan et al. (2016). Modak et al. (2018) also found negative forcing adjustments from a methane perturbation including shortwave absorption in the NCAR CAM5 model, in agreement with the above assessment. The uncertainty in the shortwave component leads to a higher radiative modelling uncertainty (14%) than for CO2 (Etminan et al., 2016). When combined with the uncertainty in the adjustment, this gives an overall uncertainty of ±20%. There is high confidence in the spectroscopic revision but only medium confidence in the adjustment modification. CH4 concentrations have increased from 729 ppb in 1750 to 1866 ppb in 2019 Section 2.2.3.3). The historical ERF estimate from AR5 of 0.48 ± 0.10 W m–2(1750–2011) is revised to 0.54 ± 0.11 W m–2(1750 to 2019) in this assessment from a combination of spectroscopic radiative efficiency revisions (+0.12 W m–2), adjustments (–0.08 W m–2) and the 63 ppb rise in atmospheric CH4 concentrations between 2011 and 2019 (+0.03 W m–2). As the adjustments are assessed to be small, there is high confidence in the overall assessment of ERF from methane. Increased methane leads to tropospheric ozone production and increased stratospheric water vapour, so that an attribution of forcing to methane emissions gives a larger effect than that directly from the methane concentration itself. This is discussed in detail in (Chapter 6 (Section 6.4.2) and shown in Figure 6.12.

7.3.2.3 Nitrous oxide (N2O)

The tropospheric adjustments to nitrous oxide (N2O) have been calculated from 5 ESMs as 7% ± 13% of the SARF (Hodnebrog et al., 2020b). This value is therefore taken as the assessed adjustment, but with low confidence. The radiative modelling uncertainty is ±10% (Etminan et al., 2016), giving an overall uncertainty of ±16%. Nitrous oxide concentrations have increased from 270 ppb in 1750 to 332 ppb in 2019 Section 2.2.3.3). The historical ERF estimate from N2O is revised upwards from 0.17 ± 0.06 W m–2(1750–2011) in AR5 to 0.21 ± 0.03 W m–2(1750–2019) in this assessment, of which 0.02 W m–2 is due to the 7 ppb increase in concentrations, and 0.02 W m–2 to the tropospheric adjustment. As the adjustments are assessed to be small there remains high confidence in the overall assessment.

Increased nitrous oxide leads to ozone depletion in the upper stratosphere which will make a positive contribution to the direct ERF here (Section 6.4.2 and Figure 6.12) when considering emissions-based estimates of ERF.

7.3.2.4 Halogenated Species

The stratospheric-temperature-adjusted radiative efficiencies (SARF per ppb increase in concentration) for halogenated compounds are reviewed extensively in Hodnebrog et al. (2020a) , an update to those used in AR5. Many halogenated compounds have lifetimes short enough that they can be considered short-lived climate forcers (SLCFs; Table 6.1). As such, they are not completely ‘well-mixed’ and their vertical distributions are taken into account when determining their radiative efficiencies. The World Meteorological Organization (WMO, 2018) updated the lifetimes of many halogenated compounds and these were used in Hodnebrog et al. (2020a) .

The tropospheric adjustments to chlorofluorocarbons (CFCs), specifically CFC-11 and CFC-12, have been quantified as 13% ± 10% and 12% ± 14% of the SARF, respectively (Hodnebrog et al., 2020b). The assessed adjustment to CFCs is therefore 12% ± 13% with low confidence due to the lack of corroborating studies. There have been no calculations for other halogenated species so for these the tropospheric adjustments are therefore assumed to be 0 ± 13% with low confidence. The radiative modelling uncertainties are 14% and 24% for compounds with lifetimes greater than and less than five years, respectively (Hodnebrog et al., 2020a). The overall uncertainty in the ERFs of halogenated compounds is therefore assessed to be 19% and 26% depending on the lifetime. The ERF from CFCs is slowly decreasing, but this is compensated for by the increased forcing from the replacement species (HCFCs and HFCs). The ERF from HFCs has increased by 0.028 ± 0.05 W m–2. Thus, the concentration changes mean that the total ERF from halogenated compounds has increased since AR5 from 0.360 ± 0.036 W m–2 to 0.408 ± 0.078 W m–2(Table 7.5). Of this, 0.034 W m–2 is due to increased radiative efficiencies and tropospheric adjustments, and 0.014 W m–2 is due to increases in concentrations. As the adjustments are assessed to be small there remains high confidence in the overall assessment.

Halogenated compounds containing chlorine and bromine lead to ozone depletion in the stratosphere which will reduce the associated ERF (Morgenstern et al., 2020). Chapter 6 (Section 6.4 and Figure 6.12) assesses the ERF contributions due to the chemical effects of reactive gases.

7.3.2.5 Ozone

Estimates of the pre-industrial to present-day tropospheric ozone radiative forcing are based entirely on models. The lack of pre-industrial ozone measurements prevents an observational determination. There have been limited studies of ozone ERFs (MacIntosh et al., 2016; Xie et al., 2016; Skeie et al., 2020). Skeie et al. (2020) found little net contribution to the ERF from tropospheric adjustment terms for 1850–2000 change in ozone (tropospheric and stratospheric ozone combined), although MacIntosh et al. (2016) suggested that increases in stratospheric or upper tropospheric ozone reduces high-cloud and increases low-cloud, whereas an increase in lower tropospheric ozone reduces low-cloud. Further studies suggest that changes in circulation due to decreases in stratospheric ozone affect Southern Hemisphere clouds and the atmospheric levels of sea salt aerosol that would contribute additional adjustments, possibly of comparable magnitude to the SARF from stratospheric ozone depletion (Grise et al., 2013, 2014; Xia et al., 2016, 2020). ESM responses to changes in ozone-depleting substances (ODS) in CMIP6 show a much more negative ERF than would be expected from offline calculations of SARF (Morgenstern et al., 2020; Thornhill et al., 2021b) again suggesting a negative contribution from adjustments. However there is insufficient evidence available to quantify this effect.

Without sufficient information to assess whether the ERFs differ from SARF, this assessment relies on offline radiative transfer calculations of SARF for both tropospheric and stratospheric ozone. Checa-Garcia et al. (2018) found SARF of 0.30 W m–2 for changes in ozone (1850–1860 to 2009–2014). These were based on precursor emissions and ODS concentrations from the Coupled Chemistry Model Initiative (CCMI) project (Morgenstern et al., 2017). Skeie et al. (2020) calculated an ozone SARF of 0.41 ± 0.12 W m–2(1850–2010; from five climate models and one chemistry transport model) using CMIP6 precursor emissions and ODS concentrations (excluding models without fully interactive ozone chemistry and one model with excessive ozone depletion). The ozone precursor emissions are higher in CMIP6 than in CCMI, which explains much of the increase compared to Checa-Garcia et al. (2018).

Previous assessments have split the ozone forcing into tropospheric and stratospheric components. This does not correspond to the division between ozone production and ozone depletion and is sensitive to the choice of tropopause (high confidence) (Myhre et al., 2013b). The contributions to total SARF in CMIP6 (Skeie et al., 2020) are 0.39 ± 0.07 and 0.02 ± 0.07 W m–2 for troposphere and stratosphere respectively (using a 150 ppb ozone tropopause definition). This small positive (but with uncertainty encompassing negative values) stratospheric ozone SARF is due to contributions from ozone precursors to lower stratospheric ozone and some of the CMIP6 models showing ozone depletion in the upper stratosphere, where depletion contributes a positive radiative forcing (medium confidence).

As there is insufficient evidence to quantify adjustments, for total ozone the assessed central estimate for ERF is assumed to be equal to SARF (low confidence) and follows Skeie et al. (2020), since that study uses the most recent emissions data. The dataset is extended over the entire historical period following Skeie et al. (2020), with a SARF for 1750–1850 of 0.03 W m–2 and for 2010–2018 of 0.03 W m–2, to give 0.47 [0.24 to 0.70] W m–2 for 1750–2019. This maintains the 50% uncertainty (5–95% range) from AR5 which is largely due to the uncertainty in pre-industrial emissions (Rowlinson et al., 2020). There is also high confidence that this range includes uncertainty due to the adjustments. The CMIP6 SARF is more positive than the AR5 value of 0.31 W m–2 for the period 1850–2011 (Myhre et al., 2013b) which was based on the Atmospheric Chemistry and Climate Intercomparison Project (ACCMIP; Shindell et al., 2013). The assessment is sensitive to the assumptions on precursor emissions used to drive the models, which are larger in CMIP6 than ACCMIP.

In summary, although there is insufficient evidence to quantify adjustments, there is high confidence in the assessed range of ERF for ozone changes over the 1750–2019 period, giving an assessed ERF of 0.47 [0.24 to 0.70] W m–2.

7.3.2.6 Stratospheric Water Vapour

This section considers direct anthropogenic effects on stratospheric water vapour by oxidation of methane. Since AR5 the SARF from methane-induced stratospheric water vapour changes has been calculated in Winterstein et al., 2019, corresponding to 0.09 W m–2 when scaling to 1850 to 2014 methane changes. This is marginally larger than the AR5 assessed value of 0.07 ± 0.05 W m–2(Myhre et al., 2013b). Wang and Huang (2020) quantified the adjustment terms to a stratospheric water vapour change equivalent to a forcing from a 2×CO2 warming (which has a different vertical profile). They found that the ERF was less than 50% of the SARF due to high-cloud decrease and upper tropospheric warming. The assessed ERF is therefore 0.05 ± 0.05 W m–2 with a lower limit reduced to zero and the central value and upper limit reduced to allow for adjustment terms. This still encompasses the two recent SARF studies. There is medium confidence in the SARF from agreement with the recent studies and AR5. There is low confidence in the adjustment terms.

Stratospheric water vapour may also change as an adjustment to species that warm or cool the upper troposphere–lower stratosphere region (Forster and Joshi, 2005; Stuber et al., 2005), in which case it should be included as part of the ERF for that compound. Changes in GSAT are also associated with changes in stratospheric water vapour as part of the water-vapour–climate feedback (Section 7.4.2.2).

7.3.2.7 Synthesis

The ERF of GHGs (excluding ozone and stratospheric water vapour) over 1750–2019 is assessed to be 3.32 ± 0.29 W m–2. It has increased by 0.49 W m–2 compared to AR5 (reference year 2011) (high confidence). Most of this has been due to an increase in CO2 concentration since 2011 [0.27 ± 0.03] W m–2, with concentration increases in CH4, N2O and halogenated compounds adding 0.02, 0.02 and 0.01 W m–2 respectively (Table 7.5). Changes in the radiative efficiencies (including adjustments) of CO2, CH4, N2O and halogenated compounds have increased the ERF by an additional 0.15 W m–2 compared to the AR5 values (high confidence). Note that the ERFs in this section do not include chemical effects of GHGs on production or destruction of ozone or aerosol formation (Section 6.2.2). The ERF for ozone is considerably increased compared to AR5 due to an increase in the assumed ozone precursor emissions in CMIP6 compared to CMIP5, and better accounting for the effects of both ozone precursors and ODSs in the stratosphere. The ERF for stratospheric water vapour is slightly reduced. The combined ERF from ozone and stratospheric water vapour has increased since AR5 by 0.10 ± 0.50 W m–2(high confidence), although the uncertainty ranges still include the AR5 values.

Table 7.5 | Present-day mole fractions in parts per trillion (pmol mol–1), except where specified, and effective radiative forcing (ERF, in W m–2) for the well-mixed greenhouse gases (WMGHGs). Data taken from (Chapter 2 Section 2.2.3). The data for 2011 (the time of the AR5 estimates) are also shown. Some of the concentrations vary slightly from those reported in AR5 owing to averaging different data sources. Individual species are reported where 1750–2019 ERF is at least 0.001 W m–2. Radiative efficiencies for the minor gases are given in Supplementary Material, Table 7.SM.7. Uncertainties in the ERF for all gases are dominated by the uncertainties in the radiative efficiencies. Tabulated global mixing ratios of all WMGHGs and ERFs from 1750 to 2019 are provided in Annex III.

Concentration

ERF with Respect to 1850

ERF with Respect to 1750

2019

2011

1850

1750

2019

2011

2019

2011

CO2 (ppm)

409.9

390.5

285.5

278.3

2.012 ± 0.241

1.738

2.156 ± 0.259

1.882

CH4(ppb)

1866.3

1803.3

807.6

729.2

0.496 ± 0.099

0.473

0.544 ± 0.109

0.521

N2O (ppb)

332.1

324.4

272.1

270.1

0.201 ± 0.030

0.177

0.208 ± 0.031

0.184

HFC-134a

107.6

62.7

0.0

0.0

0.018

0.010

0.018

0.010

HFC-23

32.4

24.1

0.0

0.0

0.006

0.005

0.006

0.005

HFC-32

20.0

4.7

0.0

0.0

0.002

0.001

0.002

0.001

HFC-125

29.4

10.3

0.0

0.0

0.007

0.002

0.007

0.002

HFC-143a

24.0

12.0

0.0

0.0

0.004

0.002

0.004

0.002

SF6

10.0

7.3

0.0

0.0

0.006

0.004

0.006

0.004

CF4

85.5

79.0

34.0

34.0

0.005

0.004

0.005

0.004

C2 f6

4.8

4.2

0.0

0.0

0.001

0.001

0.001

0.001

CFC-11

226.2

237.3

0.0

0.0

0.066

0.070

0.066

0.070

CFC-12

503.1

528.6

0.0

0.0

0.180

0.189

0.180

0.189

CFC-113

69.8

74.6

0.0

0.0

0.021

0.022

0.021

0.022

CFC-114

16.0

16.3

0.0

0.0

0.005

0.005

0.005

0.005

CFC-115

8.7

8.4

0.0

0.0

0.002

0.002

0.002

0.002

HCFC-22

246.8

213.2

0.0

0.0

0.053

0.046

0.053

0.046

HCFC-141b

24.4

21.4

0.0

0.0

0.004

0.003

0.004

0.003

HCFC-142b

22.3

21.2

0.0

0.0

0.004

0.004

0.004

0.004

CCl4

77.9

86.1

0.0

0.0

0.013

0.014

0.013

0.014

Sum of HFCs (HFC-134a equivalent)

237.1

128.6

0.0

0.0

0.040

0.022

0.040

0.022

Sum of CFCs+HCFCs+other ozone depleting gases covered by the Montreal Protocol (CFC-12 equivalent)

1031.9

1050.1

0.0

0.0

0.354

0.362

0.354

0.362

Sum of PFCs (CF4 equivalent)

109.4

98.9

34.0

34.0

0.007

0.006

0.007

0.006

Sum of Halogenated species

0.408 ±0.078

0.394

0.408 ±0.078

0.394

Total

3.118 ±0.258

2.782

3.317 ±0.278

2.981

7.3.3 Aerosols

Anthropogenic activity, and particularly burning of biomass and fossil fuels, has led to a substantial increase in emissions of aerosols and their precursors, and thus to increased atmospheric aerosol concentrations since the pre-industrial era (Sections 2.2.6 and 6.3.5, and Figure 2.9). This is particularly true for sulphate and carbonaceous aerosols (Section 6.3.5). This has in turn led to changes in the scattering and absorption of incoming solar radiation, and also affected cloud micro- and macro-physics and thus cloud radiative properties. Aerosol changes are heterogeneous in both space and time and have impacted not just Earth’s radiative energy budget but also air quality (Sections 6.1.1 and 6.6.2). Here, the assessment is focused exclusively on the global mean effects of aerosols on Earth’s energy budget, while regional changes and changes associated with individual aerosol compounds are assessed in (Chapter 6 (Sections 6.4.1 and 6.4.2).

Consistent with the terminology introduced in Box 7.1, the ERF due to changes from direct aerosol–radiation interactions (ERFari) is equal to the sum of the instantaneous top-of-atmosphere (TOA) radiation change (IRFari) and the subsequent adjustments. Likewise, the ERF following interactions between anthropogenic aerosols and clouds (ERFaci, referred to as ‘indirect aerosol effects’ in previous assessment reports) can be divided into an instantaneous forcing component (IRFaci) due to changes in cloud droplet (and indirectly also ice crystal) number concentrations and sizes, and the subsequent adjustments of cloud water content or extent. While these changes are thought to be induced primarily by changes in the abundance of cloud condensation nuclei (CCN), a change in the number of ice nucleating particles (INPs) in the atmosphere may also have occurred, and thereby contributed to ERFaci by affecting properties of mixed-phase and cirrus (ice) clouds. In the following, an assessment of IRFari and ERFari (Section 7.3.3.1) focusing on observation-based (Section 7.3.3.1.1) as well as model-based (Section 7.3.3.1.2) evidence is presented. The same lines of evidence are presented for IRFaci and ERFaci in Section 7.3.3.2. These lines of evidence are then compared with TOA energy budget constraints on the total aerosol ERf (Section 7.3.3.3) before an overall assessment of the total aerosol ERF is given in Section 7.3.3.4. For the model-based evidence, all estimates are generally valid for 2014 relative to 1750 (the time period spanned by CMIP6 historical simulations), while for observation-based evidence the assessed studies use slightly different end points, but they all generally fall within a decade (2010–2020).

7.3.3.1 Aerosol–Radiation Interactions

Since AR5, deeper understanding of the processes that govern aerosol radiative properties, and thus IRFari, has emerged. Combined with new insights into adjustments to aerosol forcing, this progress has informed new observation- and model-based estimates of ERFari and associated uncertainties.

7.3.3.1.1 Observation-based lines of evidence

Estimating IRFari requires an estimate of industrial-era changes in aerosol optical depth (AOD) and absorption AOD, which are often taken from global aerosol model simulations. Since AR5, updates to methods of estimating IRFari based on aerosol remote sensing or data-assimilated reanalyses of atmospheric composition have been published. Ma et al. (2014) applied the method of Quaas et al. (2008) to updated broadband radiative flux measurements from CERES, MODIS-retrieved AODs, and modelled anthropogenic aerosol fractions to find a clear-sky IRFari of −0.6 W m−2. This would translate into an all-sky estimate of about −0.3 W m−2 based on the clear-sky to all-sky ratio implied by Kinne (2019). Rémy et al. (2018) applied the methods of Bellouin et al. (2013a) to the reanalysis by the Copernicus Atmosphere Monitoring Service, which assimilates MODIS total AOD. Their estimate of IRFari varies between −0.5 W m–2 and −0.6 W m−2 over the period 2003–2018, and they attribute those relatively small variations to variability in biomass-burning activity. Kinne (2019) provided updated monthly total AOD and absorption AOD climatologies, obtained by blending multi-model averages with ground-based sun-photometer retrievals, to find a best estimate of IRFari of −0.4 W m−2. The updated IRFari estimates above are all scattered around the midpoint of the IRFari range of −0.35 ± 0.5 W m−2 assessed by AR5 (Boucher et al., 2013).

The more negative estimate of Rémy et al. (2018) is due to neglecting a small positive contribution from absorbing aerosols above clouds and obtaining a larger anthropogenic fraction than Kinne (2019). Rémy et al. (2018) also did not update their assumptions on black carbon anthropogenic fraction and its contribution to absorption to reflect recent downward revisions (Section 7.3.3.1.2). Kinne (2019) made those revisions, so more weight is given to that study to assess the central estimate of satellite-based IRFari to be only slightly stronger than reported in AR5 at –0.4 W m–2. While uncertainties in the anthropogenic fraction of total AOD remain, improved knowledge of anthropogenic absorption results in a slightly narrowervery likely range here than in AR5. The assessed best estimate and very likely IRFari range from observation-based evidence is therefore –0.4 ± 0.4 W m–2, but with medium confidence due to the limited number of studies available .

7.3.3.1.2 Model-based lines of evidence

While observation-based evidence can be used to estimate IRFari, global climate models are needed to calculate the associated adjustments and the resulting ERFari, using the methods described in Section 7.3.1.

A range of developments since AR5 affect model-based estimates of IRFari. Global emissions of most major aerosol compounds and their precursors are found to be higher in the current inventories, and with increasing trends. Emissions of the sulphate precursor SO2 are a notable exception; they are similar to those used in AR5 and approximately time-constant in recent decades (Hoesly et al., 2018). Myhre et al. (2017) showed, in a multi-model experiment, that the net result of these revised emissions is an IRFari trend that is relatively flat in recent years (post-2000), a finding confirmed by a single-model study by Paulot et al. (2018).

In AR5, the assessment of the black carbon (BC) contribution to IRFari was markedly strengthened in confidence by the review by Bond et al. (2013), where a key finding was a perceived model underestimate of atmospheric absorption when compared to Aeronet observations (Boucher et al., 2013). This assessment has since been revised considering: new knowledge on the effect of the temporal resolution of emissions inventories (Wang et al., 2016); the representativeness of Aeronet sites (Wang et al., 2018); issues with comparing absorption retrieval to models (E. Andrews et al., 2017); and the ageing (Peng et al., 2016), lifetime (Lund et al., 2018b) and average optical parameters (Zanatta et al., 2016) of BC. Consistent with these updates, Lund et al. (2018a) estimated the net IRFari in 2014 (relative to 1750) to be –0.17 W m–2, using CEDS emissions (Hoesly et al., 2018) as input to a chemical transport model. They attributed the weaker estimate relative to AR5 (–0.35 ± 0.5 W m–2; Myhre et al., 2013a) to stronger absorption by organic aerosol, updated parametrization of BC absorption, and slightly reduced sulphate cooling. Broadly consistent with Lund et al. (2018a) , another single-model study by Petersik et al. (2018) estimated an IRFari of –0.19 W m–2. Another single-model study by Lurton et al. (2020) reported a more negative estimate at –0.38 W m–2, but is given less weight here because the model lacked interactive aerosols and instead used prescribed climatological aerosol concentrations.

The above estimates support a less negative central estimate and a slightly narrower range compared to those reported for IRFari from ESMs in AR5 of –0.35 [–0.6 to –0.13] W m–2. The assessed central estimate and very likely IRFari range from model-based evidence alone is therefore –0.2 ± 0.2 W m–2 for 2014 relative to 1750, with medium confidence due to the limited number of studies available. Revisions due to stronger organic aerosol absorption, further developed BC parameterizations and somewhat reduced sulphate emissions in recent years.

Since AR5 considerable progress has been made in the understanding of adjustments in response to a wide range of climate forcings, as discussed in (Section 7.3.1. The adjustments in ERFari are principally caused by cloud changes, but also by lapse rate and atmospheric water vapour changes, all mainly associated with absorbing aerosols like BC. Stjern et al. (2017) found that for BC, about 30% of the (positive) IRFari is offset by adjustments of clouds (specifically, an increase in low-clouds and decrease in high-clouds) and lapse rate, by analysing simulations by five Precipitation Driver Response Model Intercomparison Project (PDRMIP) models. Smith et al. (2018b) considered more models participating in PDRMIP and suggested that about half the IRFari was offset by adjustments for BC, a finding generally supported by single-model studies (Takemura and Suzuki, 2019; Zhao and Suzuki, 2019). Thornhill et al. (2021b) also reported a negative adjustment for BC based on AerChemMIP (Collins et al., 2017) but found it to be somewhat smaller in magnitude than those reported in Smith et al. (2018b) and Stjern et al. (2017). In contrast, Allen et al. (2019) found a positive adjustment for BC and suggested that most models simulate negative adjustment for BC because of a misrepresentation of aerosol atmospheric heating profiles.

Zelinka et al. (2014) used the approximate partial radiation perturbation technique to quantify the ERFari in 2000 relative to 1860 in nine CMIP5 models; they estimated the ERFari (accounting for a small contribution from longwave radiation) to be –0.27 ± 0.35 W m–2. However, it should be noted that in Zelinka et al. (2014) adjustments of clouds caused by absorbing aerosols through changes in the thermal structure of the atmosphere (termed the semidirect effect of aerosols in AR5) are not included in ERFari but in ERFaci. The corresponding estimate emerging from the Radiative Forcing Model Intercomparison Project (RFMIP, Pincus et al., 2016) is –0.25 ± 0.40 W m–2(Smith et al., 2020b), which is generally supported by single-model studies published since AR5 (Zhang et al., 2016; Fiedler et al., 2017; Nazarenko et al., 2017; Zhou et al., 2017c, 2018b; Grandey et al., 2018). A 5% inflation is applied to the CMIP5 and CMIP6 fixed-SST derived estimates of ERFari from Zelinka et al. (2014) and Smith et al. (2020b) to account for land surface cooling (Table 7.6). Based on the above, ERFari from model-based evidence is assessed to be –0.25 ± 0.25 W m–2.

7.3.3.1.3 Overall assessment of IRFari and ERFari

The observation-based assessment of IRFari of –0.4 ± 0.4 W m–2 and the corresponding model-based assessment of –0.2 ± 0.2 W m–2 can be compared to the range of –0.45 to –0.05 W m–2 that emerged from a comprehensive review in which an observation-based estimate of anthropogenic AOD was combined with model-derived ranges for all relevant aerosol radiative properties (Bellouin et al., 2020). Based on the above, IRFari is assessed to be –0.25 ± 0.2 W m–2(medium confidence).

ERFari from model-based evidence is –0.25 ± 0.25 W m–2, which suggests a small negative adjustment relative to the model-based IRFari estimate, consistent with the literature discussed in (Section 7.3.3.1.2. Adding this small adjustment to our assessed IRFari estimate of –0.25 W m–2, and accounting for additional uncertainty in the adjustments, ERFari is assessed to –0.3 ± 0.3 (medium confidence). This assessment is consistent with the 5–95% confidence range for ERFari in Bellouin et al. (2020) of –0.71 to –0.14 W m–2, and notably implies that it is very likely that ERFari is negative. Differences relative to Bellouin et al. (2020) reflect the range of estimates in Table 7.6 and the fact that an ERFari more negative than –0.6 W m–2 would require adjustments that considerably augment the assessed IRFari, which is not supported by the assessed literature.

Table 7.6 | Present-day effective radiative forcing (ERF) due to changes in aerosol–radiation interactions (ERFari) and changes in aerosol–cloud interactions (ERFaci), and total aerosol ERF (ERFari+aci) from GCM CMIP6 (2014 relative to 1850; Smith et al., 2020b and later model results) and CMIP5 (year 2000 relative to 1860; Zelinka et al., 2014). CMIP6 results are simulated as part of RFMIP (Pincus et al., 2016). An additional 5% is applied to the CMIP5 and CMIP6 model results to account for land-surface cooling (Figure 7.4; Smith et al., 2020a).

Models

ERFari

(W m–2)

ERFaci

(W m–2)

ERFari+aci

(W m–2)

ACCESS-CM2

–0.24

–0.93

–1.17

ACCESS-ESM1-5

–0.07

–1.19

–1.25

BCC-ESM1

–0.79

–0.69

–1.48

CanESM5

–0.02

–1.09

–1.11

CESM2

+0.15

–1.65

–1.50

CNRM-CM6-1

–0.28

–0.86

–1.14

CNRM-ESM2-1

–0.15

–0.64

–0.79

EC-Earth3

–0.39

–0.50

–0.89

GFDL-CM4

–0.12

–0.72

–0.84

GFDL-ESM4

–0.06

–0.84

–0.90

GISS-E2-1-G (physics_version=1)

–0.55

–0.81

–1.36

GISS-E2-1-G (physics_version=3)

–0.64

–0.39

–1.02

HadGEM3-GC31-LL

–0.29

–0.87

–1.17

IPSL-CM6A-LR

–0.39

–0.29

–0.68

IPSL-CM6A-LR-INCA

–0.45

–0.35

–0.80

MIROC6

–0.22

–0.77

–0.99

MPI-ESM-1-2-HAM

+0.10

–1.40

–1.31

MRI-ESM2-0

–0.48

–0.74

–1.22

NorESM2-LM

–0.15

–1.08

–1.23

NorESM2-MM

–0.03

–1.26

–1.29

UKESM1-0-LL

–0.20

–0.99

–1.19

CMIP6 average and 5–95% confidence range (2014 relative to 1850)

–0.25 ± 0.40

–0.86 ± 0.57

–1.11 ± 0.38

CMIP5 average and 5–95% confidence range (2000 relative to 1860)

–0.27 ± 0.35

–0.96 ± 0.55

–1.23 ± 0.48

7.3.3.2 Aerosol–Cloud Interactions

Anthropogenic aerosol particles primarily affect water clouds by serving as additional cloud condensation nuclei (CCN) and thus increasing cloud drop number concentration (Nd; Twomey, 1959). Increasing Nd while holding liquid water content constant reduces cloud drop effective radius (re), increases the cloud albedo, and induces an instantaneous negative radiative forcing (IRFaci). The clouds are thought to subsequently adjust by a slowing of the drop coalescence rate, thereby delaying or suppressing rainfall. Rain generally reduces cloud lifetime and thereby liquid water path (LWP, i.e., the vertically integrated cloud water) and/or cloud fractional coverage (Cf; Albrecht, 1989), thus any aerosol-induced rain delay or suppression would be expected to increase LWP and/or Cf. Such adjustments could potentially lead to an ERFaci considerably larger in magnitude than the IRFaci alone. However, adding aerosols to non-precipitating clouds has been observed to have the opposite effect (i.e., a reduction in LWP and/or Cf) (Lebsock et al., 2008; Christensen and Stephens, 2011). These findings have been explained by enhanced evaporation of the smaller droplets in the aerosol-enriched environments, and resultant enhanced mixing with ambient air, leading to cloud dispersal.

A small subset of aerosols can also serve as ice nucleating particles (INPs) that initiate the ice phase in supercooled water clouds, and thereby alter cloud radiative properties and/or lifetimes. However, the ability of anthropogenic aerosols (specifically BC) to serve as INPs in mixed-phase clouds has been found to be negligible in recent laboratory studies (e.g., Vergara-Temprado et al., 2018). No assessment of the contribution to ERFaci from cloud phase changes induced by anthropogenic INPs will therefore be presented.

In ice (cirrus) clouds (cloud temperatures less than –40°C), INPs can initiate ice crystal formation at relative humidity much lower than that required for droplets to freeze spontaneously. Anthropogenic INPs can thereby influence ice crystal numbers and thus cirrus cloud radiative properties. At cirrus temperatures, certain types of BC have in fact been demonstrated to act as INPs in laboratory studies (Ullrich et al., 2017; Mahrt et al., 2018), suggesting a non-negligible anthropogenic contribution to INPs in cirrus clouds. Furthermore, anthropogenic changes to drop number also alter the number of droplets available for spontaneous freezing, thus representing a second pathway through which anthropogenic emissions could affect cirrus clouds.

7.3.3.2.1 Observation-based evidence

Since AR5, the analysis of observations to investigate aerosol–cloud interactions has progressed along several axes: (i) The framework of forcing and adjustments introduced rigorously in AR5 has helped better categorize studies; (ii) the literature assessing statistical relationships between aerosol and cloud in satellite retrievals has grown, and retrieval uncertainties are better characterized; (iii) advances have been made to infer causality in aerosol–cloud relationships.

In AR5 the statistical relationship between cloud microphysical properties and aerosol index (AI; AOD multiplied by Ångström exponent) was used to make inferences about IRFaci were assessed alongside other studies which related cloud quantities to AOD. However, it is now well-documented that the latter approach leads to low estimates of IRFaci since AOD is a poor proxy for cloud-base CCN (Penner et al., 2011; Stier, 2016). Gryspeerdt et al. (2017) demonstrated that the statistical relationship between droplet concentration and AOD leads to an inferred IRFaci that is underestimated by at least 30%, while the use of AI leads to estimates of IRFaci to within ±20%, if the anthropogenic perturbation of AI is known.

Further, studies assessed in AR5 mostly investigated linear relationships between cloud droplet concentration and aerosol (Boucher et al., 2013). Since in most cases the relationships are not linear, this leads to a bias (Gryspeerdt et al., 2016). Several studies did not relate cloud droplet concentration, but cloud droplet effective radius, to the aerosol (Brenguier et al., 2000). This is problematic because in order to infer IRFaci, stratification by cloud LWP is required (McComiskey and Feingold, 2012). Where LWP positively co-varies with aerosol retrievals (which is often the case), IRFaci inferred from such relationships is biased towards low values. Also, it is increasingly evident that different cloud regimes show different sensitivities to aerosols (Stevens and Feingold, 2009). Averaging statistics over regimes thus biases the inferred IRFaci (Gryspeerdt et al., 2014b). The AR5 concluded that IRFaci estimates tied to satellite studies generally show weak IRFaci (Boucher et al., 2013), but when correcting for the biases discussed above, this is no longer the case.

Since AR5, several studies assessed the global IRFaci from satellite observations using different methods (Table 7.7). All studies relied on statistical relationships between aerosol and cloud quantities to infer sensitivities. Four studies inferred IRFaci by estimating the anthropogenic perturbation of Nd(cloud drop number concentration). For this, Bellouin et al. (2013b) and Rémy et al. (2018) made use of regional-seasonal regressions between satellite-derived Nd and AOD following Quaas et al. (2008), while Gryspeerdt et al. (2017) used AI instead of AOD in the regression to infer IRFaci. McCoy et al. (2017b) instead used the sulphate-specific mass derived in the MERRA aerosol reanalysis that assimilated MODIS AOD (Rienecker et al., 2011). All approaches have in common the need to identify the anthropogenic perturbation of the aerosol to assess IRFaci. Gryspeerdt et al. (2017) and Rémy et al. (2018) used the same approach as Bellouin et al. (2013b), while McCoy et al. (2017b) used an anthropogenic fraction from the AEROCOM multi-model ensemble (Schulz et al., 2006). Chen et al. (2014), Christensen et al. (2016a) and Christensen et al. (2017) derived the combination of IRFaci and the LWP adjustment to IRFaci (‘intrinsic forcing’ in their terminology). They relate AI and cloud albedo statistically and use the anthropogenic aerosol fraction from Bellouin et al. (2013b). This was further refined by Hasekamp et al. (2019) who used additional polarimetric satellite information over ocean to obtain a better proxy for CCN. They derived an IRFaci of –1.14 [–1.72 to –0.84] W m–2. The variant by Christensen et al. (2017) is an update compared to the Chen et al. (2014) and Christensen et al. (2016a) studies in that it better accounts for ancillary influences on the aerosol retrievals such as aerosol swelling and three-dimensional radiative effects. McCoy et al. (2020) used the satellite-observed hemispheric difference in Nd as an emergent constraint on IRFaci as simulated by GCMs to obtain a range of –1.2 to –0.6 W m–2(95% confidence interval). Diamond et al. (2020) analysed the difference in clouds affected by ship emissions with unperturbed clouds and based on this inferred a global IRFaci of –0.69 [–0.99 to –0.44] W m–2.

Table 7. 7 | Studies quantifying aspects of the global effective radiative forcing due to aerosol–cloud interactions ERFaci that are mainly based on satellite retrievals and were published since AR5. All forcings/adjustments are presented as global annual mean values in W m–2. Most studies split the ERFaci into instantaneous radiative forcing (IRFaci) and adjustments in liquid water path (LWP) and cloud fraction (Cf) separately. All published studies only considered liquid clouds. Some studies assessed the IRFaci and the LWP adjustment together and called this ‘intrinsic forcing’ (Christensen et al., 2017) and the cloud fraction adjustment ‘extrinsic forcing’. Published uncertainty ranges are converted to 5–95% confidence intervals, and ‘n/a’ indicates that the study did not provide an estimate for the relevant IRF/ERF.

IRFaci (W m–2)

Liquid Water Path (LWP) Adjustment (W m–2)

Cloud Fraction (Cf) Adjustment (W m–2)

Reference

–0.6 ± 0.6

n/a

n/a

Bellouin et al. (2013b)

–0.4 [–0.2 to –1.0]

n/a

n/a

Gryspeerdt et al. (2017)

–1.0 ± 0.4

n/a

n/a

McCoy et al. (2017b)

n/a

n/a

–0.5 [–0.1 to –0.6]

Gryspeerdt et al. (2016)

n/a

+0.3 to 0.0

n/a

Gryspeerdt et al. (2019)

–0.8 ± 0.7

n/a

n/a

Rémy et al. (2018)

–0.53

–1.14 [–1.72 to –0.84]

–1.2 to –0.6

–0.69 [–0.99 to –0.44]

+0.15

n/a

n/a

n/a

n/a

n/a

n/a

n/a

Toll et al. (2019)

Hasekamp et al. (2019)

McCoy et al. (2020)

Diamond et al. (2020)

‘Intrinsic Forcing’

–0.5 ± 0.5

–0.5 ± 0.5

Chen et al. (2014)

–0.4 ± 0.3

n/a

Christensen et al. (2016a)

–0.3 ± 0.4

–0.4 ± 0.5

Christensen et al. (2017)

Summarizing the above findings related to statistical relationships and causal aerosol effects on cloud properties, there is high confidence that anthropogenic aerosols lead to an increase in cloud droplet concentrations. Taking the average across the studies providing IRFaci estimates discussed above and considering the general agreement among estimates (Table 7.7), IRFaci is assessed to be –0.7 ± 0.5 W m–2(medium confidence).

Multiple studies have found a positive relationship between cloud fraction and/or cloud LWP and aerosols (e.g., Nakajimaet al., 2001; Kaufman and Koren, 2006; Quaas et al., 2009). Since AR5, however, it has been documented that factors independent of causal aerosol–cloud interactions heavily influence such statistical relationships. These include the swelling of aerosols in the high relative humidity in the vicinity of clouds (Grandey et al., 2013) and the contamination of aerosol retrievals next to clouds by cloud remnants and cloud-side scattering (Várnai and Marshak, 2015; Christensen et al., 2017). Stratifying relationships by possible influencing factors such as relative humidity (Koren et al., 2010) does not yield satisfying results since observations of the relevant quantities are not available at the resolution and quality required. Another approach to tackle this problem was to assess the relationship of cloud fraction with droplet concentration (Gryspeerdt et al., 2016; Michibata et al., 2016; Sato et al., 2018). The relationship between satellite-retrieved cloud fraction and Nd was found to be positive (Christensen et al., 2016a, 2017; Gryspeerdt et al., 2016), implying an overall adjustment that leads to a more negative ERFaci. However, since retrieved Nd is biased low for broken clouds this result has been called into question (Grosvenor et al., 2018). Zhu et al. (2018) proposed to circumvent this problem by considering Nd of only continuous thick cloud covers, on the basis of which Rosenfeld et al. (2019) still obtained a positive relationship between cloud fraction and Nd relationship.

The relationship between LWP and cloud droplet number is debated. Most recent studies (primarily based on MODIS data) find negative statistical relationships (Michibata et al., 2016; Toll et al., 2017; Sato et al., 2018; Gryspeerdt et al., 2019), while Rosenfeld et al. (2019) obtained a modest positive relationship. To increase confidence that observed relationships between aerosol emissions and cloud adjustments are causal, known emissions of aerosols and aerosol precursor gases into otherwise pristine conditions have been exploited. Ship exhaust is one such source. Goren and Rosenfeld (2014) suggested that both LWP and Cf increase in response to ship emissions, contributing approximately 75% to the total ERFaci in mid-latitude stratocumulus. Christensen and Stephens (2011) found that such strong adjustments occur for open-cell stratocumulus regimes, while adjustments are comparatively small in closed-cell regimes. Volcanic emissions have been identified as another important source of information (Gassó, 2008). From satellite observations, Yuan et al. (2011) documented substantially larger Cf, higher cloud tops, reduced precipitation likelihood, and increased albedo in cumulus clouds in the plume of the Kīlauea volcano in Hawaii. Ebmeier et al. (2014) confirmed the increased LWP and albedo for other volcanoes. In contrast, for the large Holuhraun eruption in Iceland, Malavelle et al. (2017) did not find any large-scale change in LWP in satellite observations. However, when accounting for meteorological conditions, McCoy et al. (2018) concluded that for cyclonic conditions, the extra Holuhraun aerosol did enhance LWP. Toll et al. (2017) examined a large sample of volcanoes and found a distinct albedo effect, but only modest LWP changes, on average. Gryspeerdt et al. (2019) demonstrated that the negative LWP–Nd relationship becomes very small when conditioned on a volcanic eruption, and therefore concluded that LWP adjustments are small in most regions. Similarly, Toll et al. (2019) studied clouds downwind of various anthropogenic aerosol sources using satellite observations and inferred an IRFaci of –0.52 W m–2 that was partly offset by 29% due to aerosol-induced LWP decreases.

Apart from adjustments involving LWP and Cf, several studies have also documented a negative relationship between cloud-top temperature and AOD/AI in satellite observations (e.g., Koren et al., 2005). Wilcox et al. (2016) proposed that this could be explained by black-carbon (BC) absorption reducing boundary-layer turbulence, which in turn could lead to taller clouds. However, it has been demonstrated that the satellite-derived relationships are affected by spurious co-variation (Gryspeerdt et al., 2014a), and it therefore remains unclear whether a systematic causal effect exists.

Identifying relationships between INP concentrations and cloud properties from satellites is intractable because the INPs generally represent a very small subset of the overall aerosol population at any given time or location. For ice clouds, only a few satellite studies have so far investigated responses to aerosol perturbations. Gryspeerdt et al. (2018) find a positive relationship between aerosol and ice crystal number for cold cirrus under strong dynamical forcing, which could be explained by an overall larger number of solution droplets available for homogeneous freezing in polluted regions. Zhao et al. (2018) conclude that the sign of the relationship between ice crystal size and aerosol depends on humidity. While these studies support modelling results finding that ice clouds do respond to anthropogenic aerosols (Section 7.3.3.2.2), no quantitative conclusions about IRFaci or ERFaci for ice clouds can be drawn based on satellite observations.

Only a handful of studies have estimated the LWP and Cf adjustments that are needed for satellite-based estimates of ERFaci. Chen et al. (2014) and Christensen et al. (2017) used the relationship between cloud fraction and AI to infer the cloud fraction adjustment. Gryspeerdt et al. (2017) used a similar approach but tried to account for non-causal coorelations between aerosols and cloud fraction by using Ndas a mediating factor. These three studies together suggest a global Cf adjustment that augments ERFaci relative to IRFaci by –0.5 ± 0.4 W m–2(medium confidence). For global estimates of the LWP adjustment, evidence is even scarcer. Gryspeerdt et al. (2019) derived an estimate of the LWP adjustment using a method similar to Gryspeerdt et al. (2016). They estimated that the LWP adjustment offsets 0–60% of the (negative) IRFaci (0.0 to +0.3 W m–2). Supporting an offsetting LWP adjustment, Toll et al. (2019) estimated a moderate LWP adjustment of 29% (+0.15 W m–2). The adjustment due to LWP is assessed to be small, with a central estimate and very likely range of 0.2 ± 0.2 W m–2, but with low confidence due to the limited number of studies available.

Combining IRFaci and the associated adjustments in Cf and LWP (adding uncertainties in quadrature), considering only liquid-water clouds and evidence from satellite observations alone, the central estimate and very likely range for ERFaci is assessed to be –1.0 ± 0.7 W m–2(medium confidence). The confidence level and wider range for ERFaci compared to IRFaci reflect the relatively large uncertainties that remain in the adjustment contribution to ERFaci.

7.3.3.2.2 Model-based evidence

As in AR5, the representation of aerosol–cloud interactions in ESMs remains a challenge, due to the limited representation of important sub-gridscale processes, from the emissions of aerosols and their precursors to precipitation formation. ESMs that simulate ERFaci typically include aerosol–cloud interactions in liquid stratiform clouds only, while very few include aerosol interactions with mixed-phase, convective and ice clouds. Adding to the spread in model-derived estimates of ERFaci is the fact that model configurations and assumptions vary across studies, for example when it comes to the treatment of oxidants, which influence aerosol formation, and their changes through time (Karset et al., 2018).

In AR5, ERFaci was assessed as the residual of the total aerosol ERF and ERFari, as the total aerosol ERF was easier to calculate based on available model simulations (Boucher et al., 2013). The central estimates of total aerosol ERF and ERFari in AR5 were –0.9 W m–2 and –0.45 W m–2, respectively, yielding an ERFaci estimate of –0.45 W m–2. This value is much less negative than the bottom-up estimate of ERFaci from ESMs presented in AR5 (–1.4 W m–2) and efforts have been made since to reconcile this difference. Zelinka et al. (2014) estimated ERFaci to be –0.96 ± 0.55 W m–2(including semi-direct effects, and with land-surface cooling effect applied), based on nine CMIP5 models (Table 7.6). The corresponding ERFaci estimate based on 17 RFMIP models from CMIP6 is slightly less negative at –0.86 ± 0.57 W m–2(Table 7.6). Other post-AR5 estimates of ERFaci based on single-model studies are either in agreement with or slightly larger in magnitude than the CMIP6 estimate (Gordon et al., 2016; Fiedler et al., 2017, 2019; Neubauer et al., 2017; Karset et al., 2018; Regayre et al., 2018; Zhou et al., 2018b; Golaz et al., 2019; Diamond et al., 2020).

The adjustment contribution to the CMIP6 ensemble mean ERFaci is –0.20 W m–2, though with considerable differences between the models (Smith et al., 2020b). Generally, this adjustment in ESMs arises mainly from LWP changes (e.g., Ghan et al., 2016), while satellite observations suggest that cloud cover adjustments dominate and that aerosol effects on LWP are overestimated in ESMs (Bender et al., 2019). Large-eddy-simulations also tend to suggest an overestimated aerosol effect on cloud lifetime in ESMs, but some report an aerosol-induced decrease in cloud cover that is at odds with satellite observations (Seifert et al., 2015). Despite this potential disagreement when it comes to the dominant adjustment mechanism, a substantial negative contribution to ERFaci from adjustments is supported both by observational and modelling studies.

Contributions to ERFaci from anthropogenic aerosols acting as INPs are generally not included in CMIP6 models. Two global modelling studies incorporating parametrizations based on recent laboratory studies both found a negative contribution to ERFaci (Penner et al., 2018; McGraw et al., 2020), with central estimates of –0.3 and –0.13 W m–2, respectively. However, previous studies have produced model estimates of opposing signs (Storelvmo, 2017). There is thus limited evidence and medium agreement for a small negative contribution to ERFaci from anthropogenic INP-induced cirrus modifications (low confidence).

Similarly, aerosol effects on deep convective clouds are typically not incorporated in ESMs. However, cloud-resolving modelling studies support non-negligible aerosol effects on the radiative properties of convective clouds and associated detrained cloud anvils (Tao et al., 2012). While global ERF estimates are currently not available for these effects, the fact that they are missing in most ESMs adds to the uncertainty range for the model-based ERFaci.

From model-based evidence, ERFaci is assessed to –1.0 ± 0.8 W m–2(medium confidence). This assessment uses the mean ERFaci in Table 7.6 as a starting point, but further allows for a small negative ERF contribution from cirrus clouds. The uncertainty range is based on those reported in Table 7.6, but widened to account for uncertain but likely non-negligible processes currently unaccounted for in ESMs.

7.3.3.2.3 Overall assessment of ERFaci

The assessment of ERFaci based on observational evidence alone (–1.0 ± 0.7 W m–2) is very similar to the one based on model evidence alone (–1.0 ± 0.8 W m–2), in strong contrast to what was reported in AR5. This reconciliation of observation-based and model-based estimates is the result of considerable scientific progress and reflects comparable revisions of both model-based and observation-based estimates. The strong agreement between the two largely independent lines of evidence increases confidence in the overall assessment of the central estimate and very likely range for ERFaci of –1.0 ± 0.7 W m–2(medium confidence). The assessed range is consistent with but narrower than that reported by the review of Bellouin et al. (2020) of –2.65 to –0.07 W m–2. The difference is primarily due to a wider range in the adjustment contribution to ERFaci in Bellouin et al. (2020), however adjustments reported relative to IRFaci ranging from 40% to 150% in that study are fully consistent with the ERFaci assessment presented here.

7.3.3.3 Energy Budget Constraints on the Total Aerosol ERF

Energy balance models of reduced complexity have in recent years increasingly been combined with Monte Carlo approaches to provide valuable ‘top-down’ (also called inverse) observational constraints on the total aerosol ERF. These top-down approaches report ranges of aerosol ERF that are found to be consistent with the global mean temperature record and, in some cases, also observed ocean heat uptake. However, the total aerosol ERF is also used together with the historical temperature record in (Section 7.5 to constrain equilibrium climate sensitivity (ECS) and transient climate response (TCR). Using top-down estimates as a separate line of evidence also for the total aerosol ERF would therefore be circular. Nevertheless, it is useful to examine the development of these estimates since AR5, and the degree to which these estimates are consistent with the upper and lower bounds of the assessments of total aerosol ERF (ERFari+aci).

When the first top-down estimates emerged (e.g., Knutti et al., 2002), it became clear that some of the early (‘bottom-up’) ESM estimates of total aerosol ERF were inconsistent with the plausible top-down range. However, as more inverse estimates have been published, it has increasingly become clear that they too are model-dependent and span a wide range of ERF estimates, with confidence intervals that in some cases do not overlap (Forest, 2018). It has also become evident that these methods are sensitive to revised estimates of other forcings and/or updates to observational datasets. A recent review of 19 such estimates reported a mean of –0.77 W m–2 for the total aerosol ERF, and a 95% confidence interval of [–1.15 to –0.31] W m–2(Forest, 2018). Adding to that review, a more recent study using the same approach reported an estimate of total aerosol ERF of –0.89 [–1.82 to –0.01] W m–2(Skeie et al., 2018). However, in the same study, an alternative way of incorporating ocean heat content in the analysis produced a total aerosol ERF estimate of –1.34 [–2.20 to –0.46] W m–2, illustrating the sensitivity to the manner in which observations are included. A new approach to inverse estimates took advantage of independent climate radiative response estimates from eight prescribed SST and sea ice-concentration simulations over the historical period to estimate the total anthropogenic ERF. From this a total aerosol ERF of –0.8 [–1.6 to +0.1] W m–2 was derived (valid for near-present relative to the late 19th century). This range was found to be more invariant to parameter choices than earlier inverse approaches (Andrews and Forster, 2020).

Beyond the inverse estimates described above, other efforts have been made since AR5 to constrain the total aerosol ERF. For example, Stevens (2015) used a simple (one-dimensional) model to simulate the historical total aerosol ERF evolution consistent with the observed temperature record. Given the lack of temporally extensive cooling trends in the 20th-century record and the fact that the historical evolution of GHG forcing is relatively well constrained, the study concluded that a more negative total aerosol ERF than –1.0 W m–2 was incompatible with the historical temperature record. This was countered by Kretzschmar et al. (2017), who argued that the model employed in Stevens (2015) was too simplistic to account for the effect of geographical redistributions of aerosol emissions over time. Following the logic of Stevens (2015), but basing their estimates on a subset of CMIP5 models as opposed to a simplified modelling framework, Kretzschmar et al. argued that a total aerosol ERF as negative as –1.6 W m–2 was consistent with the observed temperature record. Similar arguments were put forward by Booth et al. (2018), who emphasized that the degree of non-linearity of the total aerosol ERF with aerosol emissions is a central assumption in Stevens (2015).

The historical temperature record was also the key observational constraint applied in two additional studies (Rotstayn et al., 2015; Shindell et al., 2015) based on a subset of CMIP5 models. Rotstayn et al. (2015) found a strong temporal correlation (>0.9) between the total aerosol ERF and the global surface temperature. They used this relationship to produce a best estimate for the total aerosol ERF of –0.97 W m–2, but with considerable unquantified uncertainty, in part due to uncertainties in the TCR. Shindell et al. (2015) came to a similar best estimate for the total aerosol ERF of –1.0 W m–2 and a 95% confidence interval of –1.4 to –0.6 W m–2 but based this on spatial temperature and ERF patterns in the models in comparison with observed spatial temperature patterns.

A separate observational constraint on the total ERF was proposed by Cherian et al. (2014), who compared trends in downward fluxes of solar radiation observed at surface stations across Europe (described in (Section 7.2.2.3) to those simulated by a subset of CMIP5 models. Based on the relationship between solar radiation trends and the total aerosol ERF in the models, they inferred a total aerosol ERF of –1.3 W m–2 and a standard deviation of ± 0.4 W m–2.

Based solely on energy balance considerations or other observational constraints, it is extremely likely that the total aerosol ERF is negative (high confidence), butextremely unlikely that the total aerosol ERF is more negative than –2.0 W m–2(high confidence).

7.3.3.4 Overall Assessment of Total Aerosol ERF

In AR5 (Boucher et al., 2013), the overall assessment of total aerosol ERF (ERFari+aci) used the median of all ESM estimates published prior to AR5 of –1.5 [–2.4 to –0.6] W m–2 as a starting point, but placed more confidence in a subset of models that were deemed more complete in their representation of aerosol–cloud interactions. These models, which included aerosol effects on mixed-phase, ice and/or convective clouds, produced a smaller estimate of –1.38 W m–2. Likewise, studies that constrained models with satellite observations (five in total), which produced a median estimate of –0.85 W m–2, were given extra weight. Furthermore, a longwave ERFaci of 0.2 W m–2 was added to studies that only reported shortwave ERFaci values. Finally, based on higher resolution models, doubt was raised regarding the ability of ESMs to represent the cloud-adjustment component of ERFaci with fidelity. The expert judgement was therefore that aerosol effects on cloud lifetime were too strong in the ESMs, further reducing the overall ERF estimate. The above lines of argument resulted in a total aerosol assessment of –0.9 [–1.9 to –0.1] W m–2 in AR5.

Here, the best estimate and range is revised relative to AR5 (Boucher et al., 2013), partly based on updates to the above lines of argument. Firstly, the studies that included aerosol effects on mixed-phase clouds in AR5 relied on the assumption that anthropogenic black carbon (BC) could act as INPs in these clouds, which has since been challenged by laboratory experiments (Kanji et al., 2017; Vergara-Temprado et al., 2018). There is no observational evidence of appreciable ERFs associated with aerosol effects on mixed-phase and ice clouds (Section 7.3.3.2.1), and modelling studies disagree when it comes to both their magnitude and sign (Section 7.3.3.2.2). Likewise, very few ESMs incorporate aerosol effects on deep convective clouds, and cloud-resolving modelling studies report different effects on cloud radiative properties depending on environmental conditions (Tao et al., 2012). Thus, it is not clear whether omitting such effects from ESMs would lead to any appreciable ERF biases, or if so, what the sign of such biases would be. As a result, all ESMs are given equal weight in this assessment. Furthermore, there is now a considerably expanded body of literature which suggests that early modelling studies that incorporated satellite observations may have resulted in overly conservative estimates of the magnitude of ERFaci (Section 7.3.3.2.1). Finally, based on an assessment of the longwave ERFaci in the CMIP5 models, the offset of +0.2 W m–2 applied in AR5 appears to be too large (Heyn et al., 2017). As in AR5, there is still reason to question the ability of ESMs to simulate adjustments in LWP and cloud cover in response to aerosol perturbation, but it is not clear that this will result in biases that exclusively increase the magnitude of the total aerosol ERf (Section 7.3.3.2.2).

The assessment of total aerosol ERF here uses the following lines of evidence: satellite-based evidence for IRFari; model-based evidence for IRFari and ERFari; satellite-based evidence of IRFaci and ERFaci; and finally model-based evidence for ERFaci. Based on this, ERFari and ERFaci for 2014 relative to 1750 are assessed to be –0.3 ± 0.3 W m–2 and –1.0 ± 0.7 W m–2, respectively. There is thus strong evidence for a substantive negative total aerosol ERF, which is supported by the broad agreement between observation-based and model-based lines of evidence for both ERFari and ERFaci that has emerged since AR5 (Gryspeerdt et al., 2020). However, considerable uncertainty remains, particularly with regards to the adjustment contribution to ERFaci, as well as missing processes in current ESMs, notably aerosol effects on mixed-phase, ice and convective clouds. This leads to amedium confidence in the estimate of ERFari+aci and a slight narrowing of the uncertainty range. Because the estimates informing the different lines of evidence are generally valid for approximately 2014 conditions, the total aerosol ERF assessment is considered valid for 2014 relative to 1750.

Combining the lines of evidence and adding uncertainties in quadrature, the ERFari+aci estimated for 2014 relative to 1750 is assessed to be –1.3 [–2.0 to –0.6] W m–2(medium confidence). The corresponding range from Bellouin et al. (2019) is –3.15 to –0.35 W m–2, thus there is agreement for the upper bound while the lower bound assessed here is less negative. A lower bound more negative than –2.0 W m–2 is not supported by any of the assessed lines of evidence. There is high confidence that ERFaci contributes most (75–80%) to the total aerosol effect (ERFari+aci). In contrast to AR5 (Boucher et al., 2013), it is nowvirtually certain that the total aerosol ERF is negative. Figure 7.5 depicts the aerosol ERFs from the different lines of evidence along with the overall assessments.

Figure 7.5 | Net aerosol effective radiative forcing (ERF) from different lines of evidence. The headline AR6 assessment of –1.3 [–2.0 to –0.6] W m–2 is highlighted in purple for 1750–2014 and compared to the AR5 assessment of –0.9 [–1.9 to –0.1] W m–2 for 1750–2011. The evidence comprising the AR6 assessment is shown below this: energy balance constraints [–2 to 0 W m–2 with no best estimate]; observational evidence from satellite retrievals of –1.4 [–2.2 to –0.6] W m–2; and climate model-based evidence of –1.25 [–2.1 to –0.4] W m–2. Estimates from individual CMIP5 (Zelinka et al., 2014) and CMIP6 (Smith et al., 2020b and Table 7.6) models are depicted by blue and red crosses respectively. For each line of evidence the assessed best-estimate contributions from ERFari and ERFaci are shown with darker and paler shading respectively. The observational assessment for ERFari is taken from the IRFari. Uncertainty ranges are represented by black bars for the total aerosol ERF and depictvery likely ranges. Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).

As most modelling and observational estimates of aerosol ERF have end points in 2014 or earlier, there is limited evidence available for the assessment of how aerosol ERF has changed from 2014 to 2019. However, based on a general reduction in global mean AOD over this period (Section 2.2.6 and Figure 2.9), combined with a reduction in emissions of aerosols and their precursors in updated emissions inventories (Hoesly et al., 2018), the aerosol ERF is assessed to have decreased in magnitude from about 2014 to 2019 (medium confidence). Consistent with Figure 2.10, the change in aerosol ERF from about 2014 to 2019 is assessed to be +0.2 W m–2, but with low confidence due tolimited evidence. Aerosols are therefore assessed to have contributed an ERF of –1.1 [–1.7 to –0.4] W m–2 over 1750–2019 (medium confidence).

7.3.4 Other Agents

In addition to the large anthropogenic ERFs associated with WMGHGs and atmospheric aerosols assessed in Sections 7.3.2 and 7.3.3, land-use change, contrails and aviation-induced cirrus, and light-absorbing particles deposited on snow and ice have also contributed to the overall anthropogenic ERF and are assessed in Sections 7.3.4.1, 7.3.4.2 and 7.3.4.3. Changes in solar irradiance, galactic cosmic rays, and volcanic eruptions since pre-industrial times combined represent the natural contribution to the total (anthropogenic + natural) ERF and are discussed in Sections 7.3.4.4, 7.3.4.5 and 7.3.4.6.

7.3.4.1 Land Use

Land-use forcing is defined as those changes in land-surface properties directly caused by human activity rather than by climate processes (see also Section 2.2.7). Land-use change affects the surface albedo. For example, deforestation typically replaces darker forested areas with brighter cropland, and thus imposes a negative radiative forcing on climate, while afforestation and reforestation can have the opposite effect. Precise changes depend on the nature of the forest, crops and underlying soil. Land-use change also affects the amount of water transpired by vegetation (Devaraju et al., 2015). Irrigation of land directly affects evaporation (Sherwood et al., 2018), causing a global increase of 32,500 m3s−1due to human activity. Changes in evaporation and transpiration affect the latent heat budget, but do not directly affect the top-of-atmosphere (TOA) radiative fluxes. The lifetime of water vapour is so short that the effect of changes in evaporation on the greenhouse contribution of water vapour are negligible (Sherwood et al., 2018). However, evaporation can affect the ERF through adjustments, particularly through changes in low-cloud amounts. Land management affects the emissions or removal of GHGs from the atmosphere (such as CO2, CH4, N2O). These emissions changes have the greatest effect on climate (Ward et al., 2014), however they are already included in GHG inventories. Land-use change also affects the emissions of dust and biogenic volatile organic compounds (BVOCs), which form aerosols and affect the atmospheric concentrations of ozone and methane (Section 6.2.2). The effects of land use on surface temperature and hydrology were recently assessed in SRCCL (Jia et al., 2019).

Using the definition of ERF from (Section 7.1, the adjustment in land-surface temperature is excluded from the definition of ERF, but changes in vegetation and snow cover (resulting from land-use change) are included (Boisier et al., 2013). Land-use change in the mid-latitudes induces a substantial amplifying adjustment in snow cover. Few climate model studies have attempted to quantify the ERF of land-use change. T. Andrews et al. (2017) calculated a very large surface albedo ERF (–0.47 W m–2) from 1860 to 2005 in the HadGEM2-ES model, although they did not separate out the surface albedo change from snow cover change. HadGEM2-ES is known to overestimate the amount of boreal trees and shrubs in the unperturbed state (Collins et al., 2011) so will tend to overestimate the ERF associated with land-use change. The increases in dust in HadGEM2-ES contributed an extra –0.25 W m–2, whereas cloud cover changes added a small positive adjustment (0.15 W m–2) consistent with a reduction in transpiration. A multi-model quantification of land-use forcing in CMIP6 models (excluding one outlier) (Smith et al., 2020b) found an IRF of –0.15 ± 0.12 W m–2(1850–2014), and an ERF (correcting for land-surface temperature change) of –0.11 ± 0.09 W m–2. This shows a small positive adjustment term (mainly from a reduction in cloud cover). CMIP5 models show an IRF of –0.11 [–0.16 to –0.04] W m–2(1850–2000) after excluding unrealistic models (Lejeune et al., 2020).

The contribution of land-use change to albedo changes has recently been investigated using MODIS and AVHRR to attribute surface albedo to geographically specific land-cover types (Ghimire et al., 2014). When combined with a historical land-use map (Hurtt et al., 2011) this gives a SARF of –0.15 ± 0.01 W m–2 for the period 1700–2005, of which approximately –0.12 W m–2 is from 1850. This study accounted for correlations between vegetation type and snow cover, but not the adjustment in snow cover identified in T. Andrews et al. (2017).

The indirect contributions of land-use change through biogenic emissions is very uncertain. Decreases in BVOCs reduce ozone and methane (Unger, 2014), but also reduce the formation of organic aerosols and their effects on clouds (Scott et al., 2017). Adjustments through changes in aerosols and chemistry are model dependent (Zhu et al., 2019b; Zhu and Penner, 2020), and it is not yet possible to make an assessment based on a limited number of studies.

The contribution of irrigation (mainly to low-cloud amount) is assessed as –0.05[–0.1 to 0.05] W m–2 for the historical period (Sherwood et al., 2018).

Because the CMIP5 and CMIP6 modelling studies are in agreement with Ghimire et al. (2014), that study is used as the assessed albedo ERF. Adding the irrigation effect to this gives an overall assessment of the ERF from land-use change of –0.20 ± 0.10 W m–2(medium confidence). Changes in ERF since 2014 are assumed to be small compared to the uncertainty, so this ERF applies to the period 1750–2019. The uncertainty range includes uncertainties in the adjustments.

7.3.4.2 Contrails and Aviation-induced Cirrus

ERF from contrails and aviation-induced cirrus is taken from the assessment of Lee et al. (2020), at 0.057 [0.019 to 0.098] W m–2 in 2018 (see Section 6.6.2 for an assessment of the total effects of aviation). This is rounded up to address its low confidence and the extra year of air traffic to give an assessed ERF over 1750–2019 of 0.06 [0.02 to 0.10] W m–2. This assessment is given low confidence due to the potential that processes missing from the assessment would affect the magnitude of contrails and aviation-induced cirrus ERF.

7.3.4.3 Light-absorbing Particles on Snow and Ice

In AR5, it was assessed that the effects of light-absorbing particles (LAPs) did probably not significantly contribute to recent reductions in Arctic ice and snow (Vaughan et al., 2013). The SARF from LAPs on snow and ice was assessed to 0.04 [0.02 to 0.09] W m–2(Boucher et al., 2013), a range appreciably lower than the estimates given in AR4 (Forster et al., 2007). This effect was assessed to be low confidence (medium evidence, low agreement) (Table 8.5 in Myhre et al., 2013b).

Since AR5 there has been progress in the understanding of the physical state and processes in snow that govern the albedo reduction by black carbon (BC). The SROCC (IPCC, 2019a) assessed that there is high confidence that darkening of snow by deposition of BC and other light-absorbing aerosol species increases the rate of snow melt (Section 2.2 in Hock et al., 2019; Section 3.4 in Meredith et al., 2019). C. He et al. (2018) found that taking into account both the non-spherical shape of snow grains and internal mixing of BC in snow significantly altered the effects of BC on snow albedo. The reductions of snow albedo by dust and BC have been measured and characterized in the Arctic, the Tibetan Plateau, and mid-latitude regions subject to seasonal snowfall, including North America and northern and eastern Asia (Qian et al., 2015).

Since AR5, two further studies of global IRF from black carbon on snow deposition are available, with best estimates of 0.01 W m–2(Lin et al., 2014) and 0.045 W m–2(Namazi et al., 2015). Organic carbon deposition on snow and icehas been estimated to contribute a small positive IRF of 0.001 to 0.003 W m–2(Lin et al., 2014). No comprehensive global assessments of mineral dust deposition on snow are available, although the effects are potentially large in relation to the total effect of LAPs on snow and ice forcing (Yasunari et al., 2015).

Most radiative forcing estimates have a regional emphasis. The regional focus makes estimating a global mean radiative forcing from aggregating different studies challenging, and the relative importance of each region is expected to change if the global pattern of emissions sources changes (Bauer et al., 2013). The lower bound of the assessed range of BC on snow and ice is extended to zero to encompass Lin et al. (2014), with the best estimate unchanged, resulting in 0.04 [0.00 to 0.09] W m–2. The efficacy of BC on snow forcing was estimated to be 2 to 4 times as large as for an equivalent CO2 forcing as the effects are concentrated at high latitudes in the cryosphere (Bond et al., 2013). However, it is unclear how much of this effect is due to radiative adjustments leading to a higher ERF, and how much comes from a less negative feedback α due to the high-latitude nature of the forcing. To estimate the overall ERF, the IRF is doubled assuming that part of the increased efficacy is due to adjustments. This gives an overall assessed ERF of +0.08 [0.00 to 0.18] W m–2, with low confidence.

7.3.4.4 Solar

Variations in the total solar irradiance (TSI) represent a natural external forcing agent. The dominant cycle is the solar 11-year activity cycle, which is superimposed on longer cycles (Section 2.2). Over the last three 11-year cycles, the peak-to-trough amplitude in TSI has differed by about 1 W m–2 between solar maxima and minima (Figure 2.2).

The fractional variability in the solar irradiance, over the solar cycle and between solar cycles, is much greater at short wavelengths in the 200–400 nanometre (nm) band than for the broad visible/infrared band that dominates TSI (Krivova et al., 2006). The IRF can be derived simply by ΔTSI× (1 – albedo)/4 irrespective of wavelength, where the best estimate of the planetary albedo is usually taken to be 0.29 and ΔTSIrepresents the change in total solar irradiance (Stephens et al., 2015). (The factor 4 arises because TSI is per unit area of Earth cross section presented to the Sun and IRF is per unit area of Earth’s surface). The adjustments are expected to be wavelength dependent. Gray et al. (2009) determined a stratospheric temperature adjustment of –22% to spectrally resolved changes in the solar radiance over one solar cycle. This negative adjustment is due to stratospheric heating from increased absorption by ozone at the short wavelengths, increasing the outgoing longwave radiation to space. A multi-model comparison (Smith et al., 2018b) calculated adjustments of –4% due to stratospheric temperatures and –6% due to tropospheric processes (mostly clouds), for a change in TSI across the spectrum (Figure 7.4). The smaller magnitude of the stratospheric temperature adjustment is consistent with the broad spectral change rather than the shorter wavelengths characteristic of solar variation. A single-model study also found an adjustment that acts to reduce the forcing (Modak et al., 2016). While there has not yet been a calculation based on the appropriate spectral change, the –6% tropospheric adjustment from Smith et al. (2018b) is adopted along with the Gray et al. (2009) stratospheric temperature adjustment. The ERF due to solar variability over the historical period is therefore represented by 0.72 × ΔTSI× (1 – albedo)/4 using the TSI timeseries from (Chapter 2 Section 2.2.1).

The AR5 (Myhre et al., 2013b) assessed solar SARF from around 1750 to 2011 to be 0.05 [0.00 to 0.10] W m–2 which was computed from the seven-year mean around the solar minima in 1745 (being closest to 1750) and 2008 (being the most recent solar minimum). The inclusion of tropospheric adjustments that reduce ERF (compared to SARF in AR5) has a negligible effect on the overall forcing. Prior to the satellite era, proxy records are used to reconstruct historical solar activity. In AR5, historical records were constructed using observations of solar magnetic features. In this assessment historical time series are constructed from radiogenic compounds in the biosphere and in ice cores that are formed from cosmic rays (Steinhilber et al., 2012).

In this assessment the TSI from the Paleoclimate Model Intercomparison Project Phase 4 (PMIP4) reconstruction is used (Section 2.2.1; Jungclaus et al., 2017). Proxies constructed from the14C and10Be radiogenic records for the SATIRE-M model (Vieira et al., 2011) and14C record for the PMOD model (Shapiro et al., 2011) for the 1745 solar minimum provide ERFs for 1745–2008 of –0.01, –0.02 and 0.00 W m–2 respectively. An independent dataset from the National Oceanic and Atmospheric Administration’s Climate Data Record (Coddington et al., 2016; Lean, 2018) provides an ERF for 1745–2008 of +0.03 W m–2. One substantially higher ERF estimate of +0.35 W m–2 derived from TSI reconstructions is provided by Egorova et al. (2018). However, the estimate from Egorova et al. (2018) hinges on assumptions about long-term changes in the quiet Sun for which there is no observed evidence. Lockwood and Ball (2020) analysed the relationship between observed changes in cosmic ray fluxes and recent, more accurate, TSI data and derived ERF between –0.01 and +0.02 W m–2, and Yeo et al. (2020) modelling showed the maximum possible ERF to be 0.26 ± 0.09 W m–2. Hence the Egorova et al. (2018) estimate is not explicitly taken into account in the assessment presented in this section.

In contrast to AR5, the solar ERF in this assessment uses full solar cycles rather than solar minima. The pre-industrial TSI is defined as the mean from all complete solar cycles from the start of the14C SATIRE-M proxy record in 6755 BCE to 1744 CE. The mean TSI from solar cycle 24 (2009–2019) is adopted as the assessment period for 2019. The best estimate solar ERF is assessed to be 0.01 W m–2, using the14C reconstruction from SATIRE-M, with a likely range of –0.06 to +0.08 W m–2(medium confidence). The uncertainty range is adopted from the evaluation of Lockwood and Ball (2020) using a Monte Carlo analysis of solar activity from the Maunder Minimum to 2019 from several datasets, leading to an ERF of –0.12 to +0.15 W m–2. The Lockwood and Ball (2020) full uncertainty range is halved as the period of reduced solar activity in the Maunder Minimum had ended by 1750 (medium confidence).

7.3.4.5 Galactic Cosmic Rays

Variations in the flux of galactic cosmic rays (GCR) reaching the atmosphere are modulated by solar activity and affect new particle formation in the atmosphere through their link to ionization of the troposphere (Lee et al., 2019). It has been suggested that periods of high GCR flux correlate with increased aerosol and CCN concentrations and therefore also with cloud properties (e.g., Dickinson, 1975; Kirkby, 2007).

Since AR5, the link between GCR and new particle formation has been more thoroughly studied, particularly by experiments in the CERN CLOUD chamber (Cosmics Leaving OUtdoor Droplets; Dunne et al., 2016; Kirkby et al., 2016; Pierce, 2017). By linking the GCR-induced new particle formation from CLOUD experiments to CCN, Gordon et al. (2017) found that the CCN concentration for low-clouds differed by 0.2–0.3% between solar maximum and solar minimum. Combined with relatively small variations in the atmospheric ion concentration over centennial time scales (Usoskin et al., 2015), it is therefore unlikely that cosmic ray intensity affects present-day climate via nucleation (Yu and Luo, 2014; Dunne et al., 2016; Pierce, 2017; Lee et al., 2019).

Studies continue to seek a relationship between GCR and properties of the climate system based on correlations and theory. Svensmark et al. (2017) proposed a new mechanism for ion-induced increase in aerosol growth rate and subsequent influence on the CCN concentration. The study does not include an estimate of the resulting effect on atmospheric CCN concentration and cloud radiative properties. Furthermore, Svensmark et al. (2009, 2016) find correlations between GCRs and aerosol and cloud properties in satellite and ground-based data. Multiple studies investigating this link have challenged such correlations (Kristjánsson et al., 2008; Calogovic et al., 2010; Laken, 2016).

AR5 concluded that the GCR effect on CCN is too weak to have any detectable effect on climate and no robust association was found between GCR and cloudiness (Boucher et al., 2013). Published literature since AR5 robustly supports these conclusions with key laboratory, theoretical and observational evidence. There is high confidence that GCRs contribute a negligible ERF over the period 1750–2019.

7.3.4.6 Volcanic Aerosols

There is large episodic negative radiative forcing associated with sulphur dioxide (SO­­2) being ejected into the stratosphere from explosive volcanic eruptions, accompanied by more frequent smaller eruptions (Figure 2.2 and Cross-Chapter Box 4.1). From SO2 gas, reflective sulphate aerosol is formed in the stratosphere where it may persist for months to years, reducing the incoming solar radiation. The volcanic SARF in AR5 (Myhre et al., 2013b) was derived by scaling the stratospheric aerosol optical depth (SAOD) by a factor of –25 W m–2 per unit SAOD from Hansen et al. (2005b). Quantification of the adjustments to SAOD perturbations from climate model simulations have determined a significant positive adjustment driven by a reduction in cloud amount (Figure 7.4; Marshall et al., 2020). Analysis of CMIP5 models provides a mean ERF of –20 W m–2 per unit SAOD (Larson and Portmann, 2016). Single-model studies with successive generations of Hadley Centre climate models produce estimates between –17 and –19 W m–2 per unit SAOD (Gregory et al., 2016; Marshall et al., 2020), with some evidence that ERF may be non-linear with SAOD for large eruptions (Marshall et al., 2020). Analysis of the volcanically active periods of 1982–1985 and 1990–1994 using the CESM1(WACCM) aerosol–climate model provided an SAOD-to-ERF relationship of –21.5 (± 1.1) W m–2 per unit SAOD (Schmidt et al., 2018). Volcanic SO2 emissions may contribute a positive forcing through effects on upper tropospheric ice clouds, due to additional ice nucleation on volcanic sulphate particles (Friberg et al., 2015; Schmidt et al., 2018), although one observational study found no significant effect (Meyer et al., 2015). Due to low agreement, the contribution of sulphate aerosol effects on ice clouds to volcanic ERF is not included in the overall assessment.

Non-explosive volcanic eruptions generally yield negligible global ERFs due to the short atmospheric lifetimes (a few weeks) of volcanic aerosols in the troposphere. However, as discussed in (Section 7.3.3.2, the massive fissure eruption in Holuhraun, Iceland persisted for months in 2014 and 2015 and did in fact result in a marked and persistent reduction in cloud droplet radii and a corresponding increase in cloud albedo regionally (Malavelle et al., 2017). This shows that non-explosive fissure eruptions can lead to strong regional and even global ERFs, but because the Holuhraun eruption occurred in Northern Hemisphere winter, solar insolation was weak and the observed albedo changes therefore did not result in an appreciable global ERF (Gettelman et al., 2015).

The ERF for volcanic stratospheric aerosols is assessed to be –20 ± 5 W m–2 per unit SAOD (medium confidence) based on the CMIP5 multi-model mean from the Larson and Portmann (2016) SAOD forcing efficiency calculations combined with the single-model results of Gregory et al. (2016), Schmidt et al. (2018) and Marshall et al. (2020). This is applied to the SAOD time series from (Chapter 2 Section 2.2.2) to generate a time series of ERF and temperature response shown in (Chapter 2 (Figure 2.2 and Figure 7.8, respectively). The period from 500 BCE to 1749 CE, spanning back to the start of the record of Toohey and Sigl (2017), is defined as the pre-industrial baseline and the volcanic ERF is calculated using an SAOD anomaly from this long-term mean. As in AR5, a pre-industrial to present-day ERF assessment is not provided due to the episodic nature of volcanic eruptions.

7.3.5 Synthesis of Global Mean Radiative Forcing, Past and Future

7.3.5.1 Major Changes in Forcing since the IPCC Fifth Assessment Report

The AR5 introduced the concept of effective radiative forcing (ERF) and radiative adjustments, and made a preliminary assessment that the tropospheric adjustments were zero for all species other than the effects of aerosol–cloud interaction and black carbon. Since AR5, new studies have allowed for a tentative assessment of values for tropospheric adjustments to CO2, CH4, N2O, some CFCs, solar forcing, and stratospheric aerosols, and to place a tighter constraint on adjustments from aerosol–cloud interaction (Sections 7.3.2, 7.3.3 and 7.3.4). In AR6, the definition of ERF explicitly removes the land-surface temperature change as part of the forcing, in contrast to AR5 where only sea surface temperatures were fixed. The ERF is assessed to be a better predictor of modelled equilibrium temperature change (i.e., less variation in feedback parameter) than SARf (Section 7.3.1).

As discussed in (Section 7.3.2, the radiative efficiencies for CO2, CH4 and N2O have been updated since AR5 (Etminan et al., 2016). There has been a small (1%) increase in the stratospheric-temperature-adjusted CO2 radiative efficiency, and a +5% tropospheric adjustment has been added. The stratospheric-temperature-adjusted radiative efficiency for CH4 is increased by approximately 25% (high confidence). The tropospheric adjustment is tentatively assessed to be –14% (low confidence). A +7% tropospheric adjustment has been added to the radiative efficiency for N2O and +12% to CFC-11 and CFC-12 (low confidence).

For aerosols there has been a convergence of model and observational estimates of aerosol forcing, and the partitioning of the total aerosol ERF has changed. Compared to AR5 a greater fraction of the ERF is assessed to come from ERFaci compared to the ERFari. It is now assessed as virtually certain that the total aerosol ERF (ERFari+aci) is negative.

7.3.5.2 Summary ERF Assessment

Figure 7.6 shows the industrial-era ERF estimates for 1750 to 2019 for the concentration change in different forcing agents. The assessed uncertainty distributions for each individual component are combined with a 100,000-member Monte Carlo simulation that samples the different distributions, assuming they are independent, to obtain the overall assessment of total present-day ERF (Supplementary Material 7.SM.1). The corresponding emissions-based ERF figure is shown in (Chapter 6 (Figure 6.12).

Figure 7.6 | Change in effective radiative forcing (ERF) from 1750 to 2019 by contributing forcing agents (carbon dioxide, other well-mixed greenhouse gases (WMGHGs), ozone, stratospheric water vapour, surface albedo, contrails and aviation-induced cirrus, aerosols, anthropogenic total, and solar). Solid bars represent best estimates, and very likely (5–95%) ranges are given by error bars. Non-CO2 WMGHGs are further broken down into contributions from methane (CH4), nitrous oxide (N2O) and halogenated compounds. Surface albedo is broken down into land-use changes and light-absorbing particles on snow and ice. Aerosols are broken down into contributions from aerosol–cloud interactions (ERFaci) and aerosol–radiation interactions (ERFari). For aerosols and solar, the 2019 single-year values are given (Table 7.8), which differ from the headline assessments in both cases. Volcanic forcing is not shown due to the episodic nature of volcanic eruptions. Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).

The total anthropogenic ERF over the industrial era (1750–2019) is estimated as 2.72 [1.96 to 3.48] W m–2(high confidence) (Table 7.8 and Annex III). This represents a 0.43 W m–2 increase over the assessment made in AR5 (Myhre et al., 2013b) for the period 1750–2011. This increase is a result of compensating effects. Atmospheric concentration increases of GHGs since 2011 and upwards revisions of their forcing estimates have led to a 0.59 W m–2 increase in their ERF. However, the total aerosol ERF is assessed to be more negative compared to AR5, due to revised estimates rather than trends (high confidence).

Table 7.8 | Summary table of effective radiative forcing (ERF) estimates for AR6 and comparison with the four previous IPCC assessment reports. Prior to AR5 values are stratospheric-temperature-adjusted radiative forcing (SARF). For AR5 aerosol–radiation interactions (ari) and aerosol–cloud interactions (aci) are ERF; all other values assume ERF equals SARF. Ranges shown are 5–95%. Volcanic ERF is not added to the table due to the episodic nature of volcanic eruptions which makes it difficult to compare to the other forcing mechanisms. Solar ERF is based on total solar irradiance (TSI) and not spectral variation.

Driver

Global Mean Effective Radiative Forcing (W m–2)

SAR

(1750–1993)

TAR

(1750–1998)

AR4

(1750–2005)

AR5

(1750–2011)

AR6

(1750–2019)

Comment

CO2

1.56 [1.33 to 1.79]

1.46 [1.31 to 1.61]

1.66 [1.49 to 1.83]

1.82 [1.63 to 2.01]

2.16 [1.90 to 2.41]

Increases in concentrations. Changes to radiative efficiencies.

Inclusion of tropospheric adjustments.

CH4

0.47 [0.40 to 0.54

0.48 [0.41 to 0.55]

0.48 [0.43 to 0.53]

0.48 [0.43 to 0.53]

0.54 [0.43 to 0.65]

N2O

0.14 [0.12 to 0.16]

0.15 [0.14 to 0.16]

0.16 [0.14 to 0.18]

0.17 [0.14 to 0.20]

0.21 [0.18 to 0.24]

Halogenated species

0.26 [0.22 to 0.30]

0.36 [0.31 to 0.41]

0.33 [0.30 to 0.36]

0.36 [0.32 to 0.40]

0.41 [0.33 to 0.49]

Tropospheric ozone

0.4 [0.2 to 0.6]

0.35 [0.20 to 0.50]

0.35 [0.25 to 0.65]

0.40 [0.20 to 0.60]

0.47 [0.24 to 0.71]

Revised precursor emissions. No tropospheric adjustment assessed. No troposphere–stratosphere separation.

Stratospheric ozone

–0.1 [–0.2 to –0.05]

–0.15 [–0.25 to –0.05]

–0.05 [–0.15 to 0.05]

–0.05 [–0.15 to 0.05]

Stratospheric water vapour

Not estimated

[0.01 to 0.03]

0.07 [0.02 to 0.1]

0.07 [0.02 to 0.12]

0.05 [0.00 to 0.10]

Downward revision due to adjustments.

Aerosol–radiation interactions

–0.5 [–0.25 to –1.0]

Not estimated

–0.50 [–0.90 to –0.10]

–0.45 [–0.95 to 0.05]

–0.22 [–0.47 to 0.04]

ERFari magnitude reduced by about 50% compared to AR5, based on agreement between observation-based and modelling-based evidence.

Aerosol–cloud interactions

[–1.5 to 0.0]

(sulphate only)

[–2.0 to 0.0]

(all aerosols)

–0.7 [–1.8 to –0.3]

(all aerosols)

–0.45 [–1.2 to 0.0]

–0.84 [–1.45 to –0.25]

ERFaci magnitude increased by about 85% compared to AR5, based on agreement between observation-based and modelling-based lines of evidence.

Land use

Not estimated

–0.2 [–0.4 to 0.0]

–0.2 [–0.4 to 0.0]

–0.15 [–0.25 to –0.05]

–0.20 [–0.30 to –0.10]

Includes irrigation.

Surface albedo (black + organic carbon aerosol on snow and ice)

Not estimated

Not estimated

0.10 [0.00 to 0.20]

0.04 [0.02 to 0.09]

0.08 [0.00 to 0.18]

Increased since AR5 to better account for temperature effects.

Combined contrails and aviation-induced cirrus

Not estimated

[0.00 to 0.04]

Not estimated

0.05 [0.02 to 0.15]

0.06 [0.02 to 0.10]

Narrower range since AR5.

Total anthropogenic

Not estimated

Not estimated

1.6 [0.6 to 2.4]

2.3 [1.1 to 3.3]

2.72 [1.96 to 3.48]

Increase due to GHGs, compensated slightly by aerosol ERFaci.

Solar irradiance

0.3 [0.1 to 0.5]

0.3 [0.1 to 0.5]

0.12 [0.06 to 0.30]

0.05 [0.0 to 0.10]

0.01 [–0.06 to 0.08]

Revised historical TSI estimates and methodology.

Greenhouse gases, including ozone and stratospheric water vapour from methane oxidation, are estimated to contribute an ERF of 3.84 [3.46 to 4.22] W m–2 over 1750–2019. Carbon dioxide continues to contribute the largest part (56 ± 16%) of this GHG ERF (high confidence).

As discussed in (Section 7.3.3, aerosols have in total contributed an ERF of –1.1 [–1.7 to –0.4] W m–2 over 1750–2019 (medium confidence). Aerosol–cloud interactions contribute approximately 75–80% of this ERF with the remainder due to aerosol–radiation interactions (Table 7.8).

For the purpose of comparing forcing changes with historical temperature change (Section 7.5.2), longer averaging periods are useful. The change in ERF from the second half of the 19th century (1850–1900) compared with a recent period (2006–2019) is +2.20 [1.53 to 2.91] W m–2, of which 1.71 [1.51 to 1.92] W m–2 is due to CO2.

7.3.5.3 Temperature Contribution of Forcing Agents

The estimated contribution of forcing agents to the 2019 global surface air temperature (GSAT) change relative to 1750 is shown in Figure 7.7. These estimates were produced using the concentration-derived ERF time series presented in Figure 2.10 and described in Supplementary Material 7.SM.1.3. The resulting GSAT changes over time are shown in Figure 7.8. The historical time series of ERFs for the WMGHGs can be derived by applying the ERF calculations of Section 7.3.2 to the observed time series of WMGHG concentrations in Chapter 2 Section 2.2.

Figure 7.7 | The contribution of forcing agents to 2019 temperature change relative to 1750 produced using the two-layer emulator (Supplementary Material 7.SM.2), constrained to assessed ranges for key climate metrics described in Cross-Chapter Box 7.1. The results are from a 2237-member ensemble. Temperature contributions are expressed for carbon dioxide, other well-mixed greenhouse gases (WMGHGs), ozone, stratospheric water vapour, surface albedo, contrails and aviation-induced cirrus, aerosols, solar, volcanic, and total. Solid bars represent best estimates, and very likely (5–95%) ranges are given by error bars. Dashed error bars show the contribution of forcing uncertainty alone, using best estimates of ECS (3.0°C), TCR (1.8°C) and two-layer model parameters representing the CMIP6 multi-model mean. Solid error bars show the combined effects of forcing and climate response uncertainty using the distribution of ECS and TCR from Tables 7.13 and 7.14, and the distribution of calibrated model parameters from 44 CMIP6 models. Non-CO2WMGHGs are further broken down into contributions from methane (CH4), nitrous oxide (N2O) and halogenated compounds. Surface albedo is broken down into land-use changes and light-absorbing particles on snow and ice. Aerosols are broken down into contributions from aerosol–cloud interactions (ERFaci) and aerosol–radiation interactions (ERFari). Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).

These ERF timeseries are combined with a two-layer emulator (Cross-Chapter Box 7.1 and Supplementary Material 7.SM.2) using a 2237-member constrained Monte Carlo sample of both forcing uncertainty (by sampling ERF ranges) and climate response (by sampling ECS, TCR and ocean heat capacity ranges). The net model warming over the historical period is matched to the assessment of historical GSAT warming from 1850–1900 to 1995–2014 of 0.85 [0.67 to 0.98] °C (Cross-Chapter Box 2.3) and ocean heat content change from 1971 to 2018 Section 7.2.2.2). Therefore the model gives the breakdown of the GSAT trend associated with different forcing mechanisms that are consistent with the overall GSAT change. The model assumes that there is no variation in feedback parameter across forcing mechanisms (Section 7.3.1) and variations in the effective feedback parameter over the historical record (Section 7.4.4). The distribution of ECS was informed bySection 7.5.5 and chosen to approximately maintain the best estimate and likely /very likely ranges assessed in that section (see also Supplementary Material 7.SM.2). The TCR has an ensemble median value of 1.81°C, in good agreement with (Section 7.5.5. Two error bars are shown in Figure 7.7. The dashed error bar shows the contribution of ERF uncertainty (as assessed in the subsections of (Section 7.3) employing the best estimate of climate response with an ECS of 3.0°C. The solid bar is the total response uncertainty using the (Section 7.5.5 assessment of ECS. The uncertainty in the historical temperature contributions ofthe different forcing agents is mostly due to uncertainties in ERF, yet for the WMGHG the uncertainty is dominated by the climate response as its ERF is relatively well known (Figure 7.7). From the assessment of emulator responses in Cross-Chapter Box 7.1, there is high confidence that calibrated emulators such as the one employed here can represent the historical GSAT change between 1850–1900 and 1995–2014 to within 5% for the best estimate and 10% for the very likely range (Supplementary Material, Table 7.SM.4). This gives high confidence in the overall assessment of GSAT change for the response to ERFs over 1750–2019 derived from the emulator.

The total human forced GSAT change from 1750 to 2019 is calculated to be 1.29 [1.00 to 1.65] °C (high confidence). Although the total emulated GSAT change has high confidence, the confidence of the individual contributions matches those given for the ERF assessment in the subsections of (Section 7.3. The calculated GSAT change is comprised of a WMGHG warming of 1.58 [1.17 to 2.17] °C (high confidence), a warming from ozone changes of 0.23 [0.11 to 0.39] °C (high confidence), and a cooling of –0.50 [–0.22 to –0.96] °C from aerosol effects (medium confidence). The aerosol cooling has considerable regional time dependence (Section 6.4.3) but has weakened slightly over the last 20 years in the global mean (Figures 2.10 and 7.8). There is also a –0.06 [–0.15 to +0.01] °C contribution from surface reflectance changes which is dominated by land-use change (medium confidence). Changes in solar and volcanic activity are assessed to have together contributed a small change of –0.02 [–0.06 to +0.02] °C since 1750 (medium confidence).

The total (anthropogenic + natural) emulated GSAT between 1850–1900 and 2010–2019 is 1.14 [0.89 to 1.45] °C, compared to the assessed GSAT of 1.06 [0.88 to 1.21] °c (Section 2.3.1 and Cross Chapter Box 2.3). The emulated response is slightly warmer than the observations and has a larger uncertainty range. As the emulated response attempts to constrain to multiple lines of evidence (Supplementary Material 7.SM.2), only one of which is GSAT, they should not necessarily be expected to exactly agree. The larger uncertainty range in the emulated GSAT compared to the observations is reflective of the uncertainties in ECS, TCR and ERF (particularly the aerosol ERF) that drive the emulator response.

The emulator gives a range of GSAT response for the period 1750 to 1850–1900 of 0.09 [0.04 to 0.14] °C from anthropogenic ERFs. These results are used as a line of evidence for the assessment of this change in (Chapter 1 (Cross-Chapter Box 1.2), which gives an overall assessment of 0.1°C [likely range –0.1 to +0.3] °C.

Figure 7.8 presents the GSAT time series using ERF time series for individual forcing agents rather than their aggregation. It shows that for most of the historical period the long time scale total GSAT trend estimate from the emulator closely follows the CO2 contribution. The GSAT estimate from non-CO2 greenhouse gas forcing (from other WMGHGs and ozone) has been approximately cancelled out in the global average by a cooling GSAT trend from aerosols. However, since 1980 the aerosol cooling trend has stabilized and may have started to reverse, so that over the last few decades the long-term warming has been occurring at a faster rate than would be expected due to CO2 alone (high confidence) (see also Sections 2.2.6 and 2.2.8). Throughout the record, but especially prior to 1930, periods of volcanic cooling dominate decadal variability. These estimates of the forced response are compared with model simulations and attributable warming estimates in (Chapter 3 Section 3.3.1).

Figure 7.8 | Attributed global surface air temperature change (GSAT) from 1750 to 2019 produced using the two-layer emulator (Supplementary Material 7.SM.2), forced with ERF derived in this chapter (displayed in Figure 2.10) and climate response constrained to assessed ranges for key climate metrics described in Cross-Chapter Box 7.1. The results shown are the medians from a 2237-member ensemble that encompasses uncertainty in forcing and climate response (year-2019 best estimates and uncertainties are shown in Figure 7.7 for several components). Temperature contributions are expressed for carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O), other well-mixed greenhouse gases (WMGHGs), ozone (O3), aerosols, and other anthropogenic forcings, as well as total anthropogenic, solar, volcanic, and total forcing. Shaded uncertainty bands showvery likely (5–95%)ranges. Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).

Cross-Chapter Box 7.1 | Physical Emulation of Earth System Models for Scenario Classification and Knowledge Integration in AR6

Contributors: Zebedee R.J. Nicholls (Australia), Malte Meinshausen (Australia/Germany), Piers Forster (United Kingdom), Kyle Armour (United States of America), Terje Berntsen (Norway), William Collins (United Kingdom), Christopher Jones (United Kingdom), Jared Lewis (Australia/New Zealand), Jochem Marotzke (Germany), Sebastian Milinski (Germany), Joeri Rogelj (United Kingdom/Belgium), Chris Smith (United Kingdom)

Climate model emulators are simple physically based models that are used to approximate large-scale climate responses of complex Earth system models (ESMs). Due to their low computational cost they can populate or span wide uncertainty ranges that ESMs cannot. They need to be calibrated to do this and, once calibrated, they can aid inter-ESM comparisons and act as ESM extrapolation tools to reflect and combine knowledge from ESMs and many other lines of evidence (Geoffroy et al., 2013a; Good et al., 2013; Smith et al., 2018a). In AR6, the term ‘climate model emulator’ (or simply ‘emulator’) is preferred over ‘simple’ or ‘reduced-complexity climate model’ to reinforce their use as specifically calibrated tools (Cross-Chapter Box 7.1, Figure 1). Nonetheless, simple physically based climate models have a long history of use in previous IPCC reports (Section 1.5.3.4). Climate model emulators can include carbon and other gas cycles and can combine uncertainties along the cause–effect chain, from emissions to temperature response. AR5 (M. Collins et al., 2013) used the MAGICC6 emulator (Meinshausen et al., 2011a) in a probabilistic setup (Meinshausen et al., 2009) to explore the uncertainty in future projections. A simple impulse response emulator (Good et al., 2011) was also used to ensure a consistent set of ESM projections could be shown across a range of scenarios. Chapter 8 in AR5 WGI (Myhre et al., 2013b) employed a two-layer emulator for quantifying global temperature-change potentials (GTP). In AR5 WGIII (Clarke et al., 2014), MAGICC6 was also used for the classification of scenarios, and in AR5 Synthesis Report (IPCC, 2014) this information was used to estimate carbon budgets. In SR1.5, two emulators were used to provide temperature projections of scenarios: the MAGICC6 model, which was used for the scenario classification, and the FaIR1.3 model (Millar et al., 2017; Smith et al., 2018a).

Cross-Chapter Box 7.1, Figure 1 | A comparison between the global surface air temperature (GSAT) response of various calibrated simple climate models, assessed ranges and Earth system models (ESMs). (a) and (b) compare the assessed historical GSAT time series (Section 2.3.1) with four multi-gas emulators calibrated to replicate numerous assessed ranges (panel (a); Cross-Chapter Box 7.1, Table 2) and also compares idealized CO2 -only concentration scenario response for one ESM (IPSL CM6A-LR) and multiple emulators which participated in RCMIP Phase 1 (Nicholls et al., 2020) calibrated to that single ESM (panel (b)). (c) and (d) compare this Report’s assessed ranges for GSAT warming (Box 4.1) under the multi-gas scenario SSP1-2.6 with the same calibrated emulators as in (a). For context, a range of CMIP6 ESM results are also shown (thin lines in (c) and open circles in (d)). Panel (b) adapted from Nicholls et al. (2020). Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).

The SR1.5 found that the physically based emulators produced different projected non-CO2 forcing and identified the largely unexplained differences between the two emulators used as a key knowledge gap (Forster et al., 2018). This led to a renewed effort to test the skill of various emulators. The Reduced Complexity Model Intercomparison Project (RCMIP; Nicholls et al., 2020) found that the latest generation of the emulators can reproduce key characteristics of the observed changes in global surface air temperature (GSAT) together with other key responses of ESMs (Cross-Chapter Box 7.1, Figure 1a). In particular, despite their reduced structural complexity, some emulators are able to replicate the non-linear aspects of ESM GSAT response over a range of scenarios. GSAT emulation has been more thoroughly explored in the literature than other types of emulation. Structural differences between emulation approaches lead to different outcomes and there are problems with emulating particular ESMs. In conclusion, there is medium confidence that emulators calibrated to single ESM runs can reproduce ESM projections of the forced GSAT response to other similar emissions scenarios to within natural variability (Meinshausen et al., 2011b; Geoffroy et al., 2013a; Dorheim et al., 2020; Nicholls et al., 2020; Tsutsui, 2020), although larger differences can remain for scenarios with very different forcing characteristics. For variables other than GSAT there has not yet been a comprehensive effort to evaluate the performance of emulators.

Application of emulators in AR6 WGI

Cross-Chapter Box 7.1 Table 1 shows the use of emulators within the WGI Report. The main use of emulation in the Report is to estimate GSAT change from effective radiative forcing (ERF) or concentration changes, where various versions of a two-layer energy budget emulator are used. The two-layer emulator is equivalent to a two-timescale impulse-response model (Supplementary Material 7.SM.2; Geoffroy et al., 2013b). Both a single configuration version and probabilistic forms are used. The emulator is an extension of the energy budget equation (Box 7.1, Equation 7.1) and allows for heat exchange between the upper- and deep-ocean layers, mimicking the ocean heat uptake that reduces the rate of surface warming under radiative forcing (Gregory, 2000; Held et al., 2010; Winton et al., 2010; Armour, 2017; Mauritsen and Pincus, 2017; Rohrschneider et al., 2019). Although the same energy budget emulator approach is used, different calibrations are employed in various sections, to serve different purposes and keep lines of evidence as independent as possible. Chapter 9 additionally employs projections of ocean heat content from the (Chapter 7 two-layer emulator to estimate the thermostatic component of future sea level rise (Section 9.6.3 and Supplementary Material 7.SM.2).

Cross-Chapter Box 7.1, Table 1 | Use of emulation within the WGI Report.

Section

Application and Emulator Type

Emulated Variables

Cross Chapter-Box 1.2

Estimate anthropogenic temperature change pre-1850, based on radiative forcing time series from Chapter 7. Uses the (Chapter 7 calibrated two-layer emulator: a two-layer energy budget emulator, probabilistically calibrated to AR6 ECS, TCR, historical warming and ocean heat uptake ranges, driven by the (Chapter 7 concentration-based ERFs.

GSAT

Section 3.3

Section 7.3

Investigation of the historical temperature response to individual forcing mechanisms to complement detection and attribution results. Uses the (Chapter 7 calibrated two-layer emulator.

GSAT

Box 4.1

Understanding the spread in GSAT increase of CMIP6 models and comparison to other assessments; assessment of contributions to projected temperature uncertainty. Uses a two-layer emulator calibrated to the (Chapter 7 ECS and TCR assessment driven by (Chapter 7 best-estimate ERFs.

GSAT

Section 4.6

Emulators used to assess differences in radiative forcing and GSAT response between RCP and SSP scenarios. Uses the (Chapter 7 ERF time series and the MAGICC7 probabilistic emissions-driven emulator for GSAT calibrated to the WGI assessment.

ERF, GSAT

Section 4.7

Emulator used for long-term GSAT projections (post-2100) to complement the small number of ESMs with data beyond 2100. Uses the MAGICC7 probabilistic emissions-driven emulator calibrated to the WGI assessment.

GSAT

Section 5.5

Estimated non-CO2 warming contributions of mitigation scenarios at the time of their net zero CO2 emissions for integration in the assessment of remaining carbon budgets. Uses the MAGICC7 probabilistic emissions-driven emulator calibrated to the WGI assessment.

GSAT

Section 6.6

Section 6.7

Estimated contributions to future warming from SLCFs across SSP scenarios based on ERF time series. Uses a single two-layer emulator configuration derived from the medians of MAGICC7 and FaIRv1.6.2 AR6 WG1 GSAT probabilistic responses and the best-estimate of ECS and TCR.

GSAT

Section 7.5

Estimating a process-based TCR from a process-based ECS. Uses a two-layer emulator in probabilistic form calibrated to process-based estimates from Chapter 7; a different calibration compared to the main (Chapter 7 emulator.

TCR

Section 7.6

Deriving emissions metrics. Uses two-layer emulator configurations derived from MAGICC7 and FaIRv1.6.2 AR6 WG1 probabilistic GSAT responses.

GTPs and their uncertainties

Section 9.6

Deriving global mean sea level projections. Uses the (Chapter 7 calibrated two-layer emulator for GSAT and ocean heat content, where GSAT drives regional statistical emulators of ice sheets and glaciers.

Sea level and ice loss

Section 11.2 and Cross-Chapter Box 11.1

Regional patterns of response are compared to global mean trends. Assessed literature includes projections with a regional pattern scaling and variability emulator.

Various regional information

Emissions-driven emulators (as opposed to ERF-driven or concentration-driven emulators) are also used in the Report. In (Chapter 4 Section 4.6) MAGICC7 is used to emulate GSAT beyond 2100 since its long-term response has been assessed to be fit-for-purpose to represent the behaviour of ESMs. In (Chapter 5 Section 5.5) MAGICC7 is used to explore the non-CO2 GSAT contribution in emissions scenarios. In (Chapter 6 and (Chapter 7 Section 7.6), two-layer model configurations are tuned to match the probabilistic GSAT responses of FaIRv1.6.2 and MAGICC7 emissions-driven emulators. For (Chapter 6 the two median values from FaIRv1.6.2 and MAGICC7 emulators are averaged and then matched to the best-estimate ECS of 3°C and TCR of 1.8°C (Tables 7.13 and 7.14) under the best-estimate ERF due to a doubling of CO2 of 3.93 W m–2(Table 7.4). For (Section 7.6 a distribution of responses is used from the two emulators to estimate uncertainties in global temperature change potentials (GTP).

Emissions-driven emulators for scenario classification in AR6 WGIII

As in AR5 and SR1.5, emissions-driven emulators are used to communicate outcomes of the physical climate science assessment and uncertainties to quantify the temperature outcome associated with different emissions scenarios. In particular, the computational efficiency of these emulators allows the analysis of a large number of multi-gas emissions scenarios in terms of multiple characteristics, e.g., year of peak temperature or 2030 emissions levels, in line with keeping global warming to below 1.5°C or 2.0°C.

Four emissions-driven emulators have been considered as tools for WGIII to explore the range of GSAT response to multiple scenarios beyond those assessed in WGI. The four emulators are CICERO-SCM (Skeie et al., 2017, 2021), FaIRv1.6.2 (Millar et al., 2017; Smith et al., 2018a), MAGICC7 (Meinshausen et al., 2009) and OSCARv3.1.1 (Gasser et al., 2017a, 2020). Each emulator’s probabilistic distribution has been calibrated to capture the relationship between emissions and GSAT change. The calibration is informed by the WGI assessed ranges of ECS, TCR, historical GSAT change, ERF, carbon cycle metrics and future warming projections under the (concentration-driven) SSP scenarios. The emulators are then provided as a tool for WGIII to perform a GSAT-based classification of mitigation scenarios consistent with the physical understanding assessed in WGI. The calibration step reduced the emulator differences identified in SR1.5. Note that evaluation of both central and range estimates of each emulator’s probabilistic projections is important to assess the fitness-for-purpose for the classification of scenarios in WGIII, based on information beyond the central estimate of GSAT warming.

MAGICC7 and FaIRv1.6.2 emissions-based emulators are able to represent the WGI assessment to within small differences (defined here as within typical rounding precisions of ±5% for central estimates and ±10% for ranges) across more than 80% of metric ranges (Cross-Chapter Box 7.1, Table 2). Both calibrated emulators are consistent with assessed ranges of ECS, historical GSAT, historical ocean heat uptake, total greenhouse gas ERF, methane ERF and the majority of the assessed SSP warming ranges. FaIRv1.6.2 also matches the assessed central value of TCRE and airborne fraction. Whereas, MAGICC7 matches the assessed TCR ranges as well as providing a closer fit to the SSP warming ranges for the lower-emissions scenarios. In the evaluation framework considered here, CICERO-SCM represents historical warming to within 2% of the assessed ranges and also represents future temperature ranges across the majority of the assessment, although it lacks the representation of the carbon cycle. In this framework, OSCARv3.1.1 is less able to represent the assessed projected GSAT ranges although it matches the range of airborne fraction estimates closely and the assessed historical GSATlikely range to within 0.5%. Despite these identified limitations, both CICERO-SCM and OSCARv3.1.1 provide additional information for evaluating the sensitivity of scenario classification to model choice.

How emulators match the assessed ranges used for the evaluation framework is summarized here and in Table 2. The first is too-low projections for 2081–2100 under SSP1-1.9 (8% or 15% too low for the central estimate and 15% or 25% too low for the lower end in the case of MAGICC7 or FaIRv1.6.2, respectively). The second is the representation of the aerosol ERF (both MAGICC7 and FaIRv1.6.2 are greater than 8% less negative than the central assessed range and greater than 10% less negative for the lower assessed range), as energy balance models struggle to reproduce an aerosol ERF with a magnitude as strong as the assessed best estimate and still match historical warming estimates. Both emulators have medium to large differences compared to the TCRE and airborne fraction ranges (see notes beneath Cross-Chapter Box 7.1, Table 2). Finally, there is also a slight overestimate of the low end of the assessed historical GSAT range.

Overall, there is high confidence that emulated historical and future ranges of GSAT change can be calibrated to be internally consistent with the assessment of key physical-climate indicators in this Report: greenhouse gas ERFs, ECS and TCR. When calibrated to match the assessed ranges of GSAT and multiple physical climate indicators, physically based emulators can reproduce the best estimate of GSAT change over 1850–1900 to 1995–2014 to within 5% and the very likely range of this GSAT change to within 10%. MAGICC7 and FaIRv1.6.2 match at least two-thirds of the (Chapter 4 assessed projected GSAT changes to within these levels of precision.

Cross-Chapter Box 7.1, Table 2 | Percentage differences between the emulator value and the WGI assessed best estimate and range for key metrics. Values are given for four emulators in their respective AR6-calibrated probabilistic setups. Absolute values of these indicators are shown in Supplementary Material, Table 7.SM.4.

Emulator

CICERO-SCM

FaIRv1.6.2

MAGICC7

OSCARv3.1.1

Assessed Range

Lower

Central

Upper

Lower

Central

Upper

Lower

Central

Upper

Lower

Central

Upper

Key metrics

ECS (°C)

26%

2%

–18%

3%

–2%

1%

–3%

–1%

–3%

–8%

–15%

–22%

TCRE (°C per 1000 GtC)**

29%

–7%

–21%

37%

5%

–5%

50%

–8%

–20%

TCR (°C)

15%

–5%

–3%

14%

0%

3%

6%

4%

9%

26%

1%

–14%

Historical warming and Effective Radiative Forcing

GSAT warming (°C)

1995–2014 rel. 1850–1900

2%

0%

0%

7%

3%

4%

7%

1%

–1%

–0%

–8%

–0%

Ocean heat content change (ZJ)*

1971–2018

–24%

–27%

–29%

5%

–4%

–9%

–1%

–3%

–6%

–47%

–39%

10%

Total Aerosol ERF (W m–2)

2005–2014 rel. 1750

36%

37%

10%

16%

12%

0%

10%

8%

8%

38%

15%

–31%

GHG ERF (W m–2)

2019 rel. 1750

4%

–5%

–13%

1%

2%

1%

2%

1%

–0%

1%

3%

–3%

Methane ERF (W m–2)

2019 rel. 1750

31%

4%

–13%

3%

3%

3%

0%

–0%

3%

8%

–1%

–5%

Carbon Cycle metrics

Airborne Fraction1pctCO2(dimensionless)*

2×CO2

8%

–3%

–11%

12%

6%

–1%

1%

–0%

8%

Airborne Fraction1pctCO2(dimensionless)*

4×CO2

12%

1%

–9%

15%

4%

–6%

5%

–1%

–1%

Future warming (GSAT) relative to 19952014

SSP1-1.9 (°C)

2021–2040

10%

–4%

10%

3%

1%

11%

2%

–0%

4%

12%

–9%

–25%

2041–2060

8%

–9%

7%

–11%

–8%

6%

–1%

–1%

7%

12%

–8%

–31%

2081–2100

–12%

–25%

–2%

–25%

–15%

4%

–15%

–8%

3%

7%

–10%

–31%

SSP1-2.6 (°C)

2021–2040

7%

–5%

5%

2%

1%

8%

–1%

–2%

–0%

9%

–9%

–28%

2041–2060

8%

–6%

2%

–2%

–2%

5%

0%

1%

2%

15%

–6%

–28%

2081–2100

–2%

–14%

–5%

–8%

–7%

1%

–6%

–1%

1%

17%

–9%

–29%

SSP2-4.5 (°C)

2021–2040

8%

–5%

5%

7%

–1%

2%

3%

–3%

–2%

–5%

–14%

–30%

2041–2060

4%

–4%

3%

1%

–1%

2%

1%

1%

2%

8%

–8%

–28%

2081–2100

–1%

–10%

–3%

–2%

–3%

1%

–2%

1%

3%

8%

–4%

–25%

SSP3-7.0 (°C)

2021–2040

11%

–4%

1%

14%

1%

–1%

10%

1%

–0%

–5%

–15%

–29%

2041–2060

4%

–5%

–0%

6%

0%

–1%

7%

4%

1%

7%

–8%

–26%

2081–2100

–0%

–8%

–3%

3%

–1%

–1%

6%

3%

6%

5%

–6%

–25%

SSP5-8.5 (°C)

2021–2040

5%

–7%

2%

9%

2%

4%

7%

1%

2%

1%

–14%

–30%

2041–2060

2%

–8%

–1%

4%

0%

4%

3%

2%

4%

10%

–6%

–24%

2081–2100

4%

–7%

–3%

6%

–0%

1%

8%

4%

7%

9%

–4%

–25%

Notes. Metrics calibrated against are equilibrium climate sensitivity, ECs (Section 7.5); transient climate response to cumulative CO2 emissions, TCRe (Section 5.5); transient climate response, TCr (Section 7.5), historical GSAT change (Section 2.3); ocean heat uptake (Sections 7.2 and 2.3); effective radiative forcing, ERf (Section 7.3); carbon cycle metrics, namely airborne fractions of idealized CO2 scenarios (taking the likely range as twice the standard deviation across the models analysed in Arora et al. (2020; see also Table 5.7, ‘cross-AR6 lines of evidence’ row); and GSAT projections under the concentration-driven SSP scenarios for the near term (2021–2040), mid-term (2041–2060) and long term (2081–2100) relative to 1995–2014 (Table 4.2). See Supplementary Material, Table 7.SM.4 for a version of this table with the absolute values rather than percentage differences. The columns labelled ‘upper’ and ‘lower’ indicate 5–95% ranges, except for the variables demarcated with an asterisk or double asterisk (* or **), where they denote likely ranges from 17–83%. Note that the TCRE assessed range (**) is wider than the combination of the TCR and airborne fraction to account for uncertainties related to model limitations (Table 5.7) hence it is expected that the emulators are too narrow on this particular metric and/or too wide on TCR and airborne fraction. For illustrative purposes, the cells are coloured as follows: white cells indicate small differences (up to ±5% for the central value and +10% for the ranges), light blue and light yellow cells indicate medium differences (up to +10% and –10% for light blue and light yellow for central values, respectively; up to ±20% for the ranges) and darker cells indicate larger positive (blue) or negative (yellow) differences. Note that values are rounded after the colours are applied.

7.4 Climate Feedbacks

The magnitude of global surface temperature change primarily depends on the strength of the radiative forcings and feedbacks, the latter defined as the changes of the net energy budget at the top-of-atmosphere (TOA) in response to a change in the GSAT (Box 7.1, Equation 7.1). Feedbacks in the Earth system are numerous, and it can be helpful to categorize them into three groups: (i) physical feedbacks; (ii) biogeophysical and biogeochemical feedbacks; and (iii) long-term feedbacks associated with ice sheets. The physical feedbacks (e.g., those associated with changes in lapse rate, water vapour, surface albedo, or clouds; (Sections 7.4.2.1–7.4.2.4) and biogeophysical/biogeochemical feedbacks (e.g., those associated with changes in methane, aerosols, ozone, or vegetation; Section 7.4.2.5) act both on time scales that are used to estimate the equilibrium climate sensitivity (ECS) in models (typically 150 years, see Box 7.1) and on longer time scales required to reach equilibrium. Long-term feedbacks associated with ice sheets (Section 7.4.2.6) are relevant primarily after several centuries or more. The feedbacks associated with biogeophysical/biogeochemical processes and ice sheets, often collectively referred to as Earth system feedbacks, had not been included in conventional estimates of the climate feedback (e.g., Hansen et al., 1984), but the former can now be quantified and included in the assessment of the total (net) climate feedback. Feedback analysis represents a formal framework for the quantification of the coupled interactions occurring within a complex Earth system in which everything influences everything else (e.g., Roe, 2009). As used here (as presented in Section 7.4.1), the primary objective of feedback analysis is to identify and understand the key processes that determine the magnitude of the surface temperature response to an external forcing. For each feedback, the basic underlying mechanisms and their assessments are presented in Section 7.4.2.

Up until AR5, process understanding and quantification of feedback mechanisms were based primarily on global climate models. Since AR5, the scientific community has undertaken a wealth of alternative approaches, including observational and fine-scale modelling approaches. This has in some cases led to more constrained feedbacks and, on the other hand, uncovered shortcomings in global climate models, which are starting to be corrected. Consequently, AR6 achieves a more robust assessment of feedbacks in the climate system that is less reliant on global climate models than in earlier assessment reports.

It has long been recognized that the magnitude of climate feedbacks can change as the climate state evolves over time (Manabe and Bryan, 1985; Murphy, 1995), but the implications for projected future warming have been investigated only recently. Since AR5, progress has been made in understanding the key mechanisms behind this time- and state-dependence. Specifically, the state-dependence is assessed by comparing climate feedbacks between warmer and colder climate states inferred from paleoclimate proxies and model simulations (Section 7.4.3). The time-dependence of the feedbacks is evident between the historical period and future projections and is assessed to arise from the evolution of the surface warming pattern related to changes in zonal and meridional temperature gradients (Section 7.4.4).

7.4.1 Methodology of the Feedback Assessment

The global surface temperature changes of the climate system are generally analysed with the classical forcing–feedback framework as described in Box 7.1 (Equation 7.1). In this equation α is the net feedback parameter (W m–2°C–1). As surface temperature changes in response to the TOA energy imbalance, many other climate variables also change, thus affecting the radiative flux at the TOA. The aggregate feedback parameter can then be decomposed into an approximate sum of terms α =Σx α x, where x is a vector representing variables that have a direct effect on the net TOA radiative flux N and

Following the conventional definition, the physical climate feedbacks are here decomposed into terms associated with a vertically uniform temperature change (Planck response, P), changes in the water-vapour plus temperature lapse-rate (WV+LR), surface albedo (A) and clouds (C). The water-vapour plus temperature lapse rate feedback is further decomposed using two different approaches, one based on changes in specific humidity, the other on changes in relative humidity. Biogeochemical feedbacks arise due to changes in aerosols and atmospheric chemical composition in response to changes in surface temperature, and Gregory et al. (2009) and Raes et al. (2010) show that they can be analysed using the same framework as for the physical climate feedbacks (Sections 5.4 and 6.4.5). Similarly, feedbacks associated with biogeophysical and ice-sheet changes can also be incorporated.

In global climate models, the feedback parameters α x in global warming conditions are often estimated as the mean differences in the radiative fluxes between atmosphere-only simulations in which the change in SST is prescribed (Cess et al., 1990), or as the regression slope of change in radiation flux against change in GSAT using atmosphere–ocean coupled simulations with abrupt CO2 changes (abrupt 4xCO2 ) for 150 years (Box 7.1; Gregory et al., 2004; Andrews et al., 2012; Caldwell et al., 2016). Neither method is perfect, but both are useful and yield consistent results (Ringer et al., 2014). In the regression method, the radiative effects of land warming are excluded from the ERF due to doubling of CO2Section 7.3.2), which may overestimate feedback values by about 15%. At the same time, the feedback calculated using the regression over years 1–150 ignores its state-dependence on multi-centennial time scales (Section 7.4.3), probably giving an underestimate of α by about 10% (Rugenstein et al., 2019). These effects are both small and approximately cancel each other in the ensemble mean, justifying the use of regression over 150 years as an approximation to feedbacks in ESMs.

The change of the TOA radiative fluxn as a function of the change of a climate variable x (such as water vapour) is commonly computed using the ‘radiative kernel’ method (Soden et al., 2008). In this method, the kernelN/x is evaluated by perturbingx within a radiation code. Then multiplying the kernel by d x/d T inferred from observations, meteorological analysis or GCMs produces a value of α x.

Feedback parameters from lines of evidence other than global models are estimated in various ways. For example, observational data combined with GCM simulations could produce an emergent constraint on a particular feedback (Hall and Qu, 2006; Klein and Hall, 2015), or the observed interannual fluctuations in the global mean TOA radiation and the surface air temperature, to which the linear regression analysis is applied, could generate a direct estimate of the climate feedback, assuming that the feedback associated with internal climate variability at short time scales can be a surrogate of the feedback to CO2 -induced warming (Dessler, 2013; Loeb et al., 2016). The assumption is not trivial, but can be justified given that the climate feedbacks are fast enough to occur at the interannual time scale. Indeed, a broad agreement has been obtained in estimates of individual physical climate feedbacks based on interannual variability and longer climate change time scales in GCMs (Zhou et al., 2015; Colman and Hanson, 2017). This means that the climate feedbacks estimated from the observed interannual fluctuations are representative of the longer-term feedbacks (decades to centuries). Care must be taken for these observational estimates because they can be sensitive to details of the calculation such as data sets and periods used (Dessler, 2013; Proistosescu et al., 2018). In particular, there would be a dependence of physical feedbacks on the surface warming pattern at the interannual time scale due, for example, to El Niño–Southern Oscillation. However, this effect both amplifies and suppresses the feedback when data include the positive and negative phases of the interannual fluctuation, and therefore the net bias will be small.

In summary, the classical forcing–feedback framework has been extended to include biogeophysical and non-CO2 biogeochemical feedbacks in addition to the physical feedbacks. It has also been used to analyse seasonal and interannual-to-decadal climate variations in observations and ESMs, in addition to long-term climate changes as seen inabrupt 4xCO2 experiments. These developments allow an assessment of the feedbacks based on a larger variety of lines of evidence compared to AR5.

7.4.2 Assessing Climate Feedbacks

This section provides an overall assessment of individual feedback parameters, α x, by combining different lines of evidence from observations, theory, process models and ESMs. To achieve this, we review the understanding of the key processes governing the feedbacks, why the feedback estimates differ among models, studies or approaches, and the extent to which these approaches yield consistent results. The individual terms assessed are the Planck response (Section 7.4.2.1) and feedbacks associated with changes in water vapour and lapse rate (Section 7.4.2.2), surface albedo (Section 7.4.2.3), clouds (Section 7.4.2.4), biogeophysical and non-CO2 biogeochemical processes (Section 7.4.2.5), and ice sheets (Section 7.4.2.6). A synthesis is provided in (Section 7.4.2.7. Climate feedbacks in CMIP6 models are then evaluated in (Section 7.4.2.8, with an explanation of how they have been incorporated into the assessment.

7.4.2.1 Planck Response

The Planck response represents the additional thermal or longwave (LW) emission to space arising from vertically uniform warming of the surface and the atmosphere. The Planck response α P, often called the Planck feedback, plays a fundamental stabilizing role in Earth’s climate and has a value that is strongly negative: a warmer planet radiates more energy to space. A crude estimate of α Pcan be made using the normalized greenhouse effect , defined as the ratio between the greenhouse effectG and the upwelling LW flux at the surface (Raval and Ramanathan, 1989). Current estimates (Section 7.2, Figure 7.2) give G= 159 W m–2 and ≈ 0.4. Assuming is constant, one obtains for a surface temperature Ts= 288 K, α P= (g– 1) 4σT3s≈ –3.3 W m–2°C–1, whereσ is the Stefan–Boltzmann constant. This parameter α P is estimated more accurately using kernels obtained from meteorological reanalysis or climate simulations (Soden and Held, 2006; Dessler, 2013; Vial et al., 2013; Caldwell et al., 2016; Colman and Hanson, 2017; Zelinka et al., 2020). Discrepancies among estimates primarily arise because differences in cloud distributions make the radiative kernels differ (Kramer et al., 2019). Using six different kernels, Zelinka et al. (2020) obtained a spread of ±0.1 W m–2°C–1(one standard deviation). Discrepancies among estimates secondarily arise from differences in the pattern of equilibrium surface temperature changes among ESMs. For the CMIP5 and CMIP6 models this introduces a spread of ±0.04 W m–2°C–1(one standard deviation). The multi-kernel and multi-model mean of α P is equal to –3.20 W m–2°C–1for the CMIP5 and –3.22 W m–2°C–1for the CMIP6 models (Supplementary Material, Table 7.SM.5). Overall, there is high confidence in the estimate of the Planck response, which is assessed to be α P= –3.22 W m–2°C–1 with avery likely range of –3.4 to –3.0 W m–2°C–1and a likely range of –3.3 to –3.1 W m–2°C–1.

The Planck temperature response ΔTP is the equilibrium temperature change in response to a forcing ΔF when the net feedback parameter is equal to the Planck response parameter: ΔTP= –ΔF / α P.

7.4.2.2 Water-vapour and Temperature Lapse-rate Feedbacks

Two decompositions are generally used to analyse the feedbacks associated with a change in the water-vapour and temperature lapse-rate in the troposphere. As in any system, many feedback decompositions are possible, each of them highlighting a particular property or aspect of the system (Ingram, 2010; Held and Shell, 2012; Dufresne and Saint-Lu, 2016). The first decomposition considers separately the changes (and therefore feedbacks) in the lapse rate (LR) and specific humidity (WV). The second decomposition considers changes in the lapse rate assuming constant relative humidity (LR*) separately from changes in relative humidity (RH).

The specific humidity (WV) feedback, also known as the water-vapour feedback, quantifies the change in radiative flux at the TOA due to changes in atmospheric water vapour concentration associated with a change in global mean surface air temperature. According to theory, observations and models, the water vapour increase approximately follows the Clausius–Clapeyron relationship at the global scale with regional differences dominated by dynamical processes (Section 8.2.1; Sherwood et al., 2010a; Chung et al., 2014; Romps, 2014; R. Liu et al., 2018; Schröder et al., 2019). Greater atmospheric water vapour content, particularly in the upper troposphere, results in enhanced absorption of LW and SW radiation and reduced outgoing radiation. This is a positive feedback. Atmospheric moistening has been detected in satellite records (Section 2.3.1.3.3), it is simulated by climate models (Section 3.3.2.2), and the estimates agree within model and observational uncertainty (Soden et al., 2005; Dessler, 2013; Gordon et al., 2013; Chung et al., 2014). The estimate of this feedback inferred from satellite observations is α WV= 1.85 ± 0.32 W m–2°C–1(R. Liu et al., 2018). This is consistent with the value α WV= 1.77 ± 0.20 W m–2°C–1(one standard deviation) obtained with CMIP5 and CMIP6 models (Zelinka et al., 2020).

The lapse-rate (LR) feedback quantifies the change in radiative flux at the TOA due to a non­uniform change in the vertical temperature profile. In the tropics, the vertical temperature profile is mainly driven by moist convection and is close to a moist adiabat. The warming is larger in the upper troposphere than in the lower troposphere (Manabe and Wetherald, 1975; Santer et al., 2005; Bony et al., 2006), leading to a larger radiative emission to space and therefore a negative feedback. This larger warming in the upper troposphere than at the surface has been observed over the last 20 years thanks to the availability of sufficiently accurate observations (Section 2.3.1.2.2). In the extratropics, the vertical temperature profile is mainly driven by a balance between radiation, meridional heat transport and ocean heat uptake (Rose et al., 2014). Strong winter temperature inversions lead to warming that is larger in the lower troposphere (Payne et al., 2015; Feldl et al., 2017a) and a positive LR feedback in polar regions (Section 7.4.4.1; Manabe and Wetherald, 1975; Bintanja et al., 2012; Pithan and Mauritsen, 2014). However, the tropical contribution dominates, leading to a negative global mean LR feedback (Soden and Held, 2006; Dessler, 2013; Vial et al., 2013; Caldwell et al., 2016). The LR feedback has been estimated at interannual time scales using meteorological reanalysis and satellite measurements of TOA fluxes (Dessler, 2013). These estimates from climate variability are consistent between observations and ESMs (Dessler, 2013; Colman and Hanson, 2017). The mean and standard deviation of this feedback under global warming based on the cited studies are α LR= –0.50 ± 0.20 W m–2°C–1(Dessler, 2013; Caldwell et al., 2016; Colman and Hanson, 2017; Zelinka et al., 2020).

The second decomposition was proposed by Held and Shell (2012) to separate the response that would occur under the assumption that relative humidity remains constant from that due to the change in relative humidity. The feedback is decomposed into three: (i) change in water vapour due to an identical temperature increase at the surface and throughout the troposphere assuming constant relative humidity, which will be called the Clausius–Clapeyron (CC) feedback here; (ii) change in LR assuming constant relative humidity (LR*); (iii) change in relative humidity (RH). Since AR5 it has been clarified that by construction, the sum of the temperature lapse rate and specific humidity (LR + WV) feedbacks is equal to the sum of the Clausius–Clapeyron feedback, the lapse rate feedback assuming constant relative humidity, and the feedback from changes in relative humidity (that is, CC + LR* + RH). Therefore, each of these two sums may simply be referred to as the ‘water-vapour plus lapse-rate’ feedback.

The CC feedback has a large positive value due to well understood thermodynamic and radiative processes: α CC= 1.36 ± 0.04 W m–2°C–1(one standard deviation; Held and Shell, 2012; Zelinka et al., 2020). The lapse-rate feedback assuming a constant relative humidity (LR*) in CMIP6 models has small absolute values ( α LR*= –0.10 ± 0.07 W m–2°C–1(one standard deviation)), as expected from theoretical arguments (Ingram, 2010, 2013). It includes the pattern effect of surface warming that modulates the lapse rate and associated specific humidity changes (Po-Chedley et al., 2018b). The relative humidity feedback is close to zero ( α RH= 0.00 ± 0.06 W m–2°C–1(one standard deviation)) and the spread among models is confined to the tropics (Sherwood et al., 2010b; Vial et al., 2013; Takahashi et al., 2016; Po-Chedley et al., 2018b). The change in upper tropospheric RH is closely related to model representation of current climate (Sherwood et al., 2010b; Po-Chedley et al., 2019), and a reduction in model RH biases is expected to reduce the uncertainty of the RH feedback. At interannual time scales, it has been shown that the change in RH in the tropics is related to the change of the spatial organization of deep convection (Holloway et al., 2017; Bony et al., 2020).

Both decompositions allow estimates of the sum of the lapse-rate and specific humidity feedbacks α LR+WV. The multi-kernel and multi-model mean of α LR+WV is equal to 1.24 and 1.26 W m–2°C–1respectively for CMIP5 and CMIP6 models, with a standard deviation of 0.10 W m–2°C–1(Zelinka et al., 2020). These values are larger than the recently assessed value of 1.15 W m–2°C–1by Sherwood et al. (2020) as a larger set of kernels, including those obtained from meteorological reanalysis, are used here.

Since AR5, the effect of the water vapour increase in the stratosphere as a result of global warming has been investigated by different studies. This increase produces a positive feedback between 0.1 and 0.3 W m–2°C–1if the stratospheric radiative response is computed assuming temperatures that are adjusted with fixed dynamical heating (Dessler et al., 2013; Banerjee et al., 2019). However, various feedbacks reduce this temperature adjustment and the overall physical (water vapour, temperature and dynamical) stratospheric feedback becomes much smaller (0.0 to 0.1 W m–2°C–1; Huang et al., 2016, 2020; Li and Newman, 2020), with uncertainty arising from limitations of current ESMs in simulating stratospheric processes. The total stratospheric feedback is assessed at 0.05 ± 0.1 W m–2°C–1(one standard deviation).

The combined ‘water-vapour plus lapse-rate’ feedback is positive. The main physical processes that drive this feedback are well understood and supported by multiple lines of evidence including models, theory and observations. The combined ‘water-vapour plus lapse-rate’ feedback parameter is assessed to be α LR+WV= 1.30 W m–2°C–1, with avery likely range of 1.1 to 1.5 W m–2°C–1and a likely range of 1.2 to 1.4 W m–2°C–1 with high confidence.

7.4.2.3 Surface-albedo Feedback

Surface albedo is determined primarily by reflectance at Earth’s surface, but also by the spectral and angular distribution of incident solar radiation. Changes in surface albedo result in changes in planetary albedo that are roughly reduced by two-thirds, owing to atmospheric absorption and scattering, with variability and uncertainty arising primarily from clouds (Bender, 2011; Donohoe and Battisti, 2011; Block and Mauritsen, 2013). Temperature change induces surface-albedo change through several direct and indirect means. In the present climate and at multi-decadal time scales, the largest contributions by far are changes in the extent of sea ice and seasonal snow cover, as these media are highly reflective and are located in regions that are close to the melting temperature (Sections 2.3.2.1 and 2.3.2.2). Reduced snow cover on sea ice may contribute as much to albedo feedback as reduced extent of sea ice (Zhang et al., 2019). Changes in the snow metamorphic rate, which generally reduces snow albedo with warmer temperature, and warming-induced consolidation of light-absorbing impurities near the surface, also contribute secondarily to the albedo feedback (Flanner and Zender, 2006; Qu and Hall, 2007; Doherty et al., 2013; Tuzet et al., 2017). Other contributors to albedo change include vegetation state (assessed separately in (Section 7.4.2.5), soil wetness and ocean roughness.

Several studies have attempted to derive surface-albedo feedback from observations of multi-decadal changes in climate, but only over limited spatial and inconsistent temporal domains, inhibiting a purely observational synthesis of global surface-albedo feedback ( α A). Flanner et al. (2011) applied satellite observations to determine that the northern hemisphere (NH) cryosphere contribution to global α Aover the period 1979–2008 was 0.48 [likely range 0.29 to 0.78] W m–2°C–1, with roughly equal contributions from changes in land snow cover and sea ice. Since AR5, and over similar periods of observation, Crook and Forster (2014) found an estimate of 0.8 ± 0.3 W m–2°C–1(one standard deviation) for the total NH extratropical surface-albedo feedback, when averaged over global surface area. For Arctic sea ice alone, Pistone et al. (2014) and Cao et al. (2015) estimated the contribution to global α Ato be 0.31 ± 0.04 W m–2°C–1(one standard deviation) and 0.31 ± 0.08 W m–2°C–1(one standard deviation), respectively, whereas Donohoe et al. (2020) estimated it to be only 0.16 ± 0.04 W m–2°C–1(one standard deviation). Much of this discrepancy can be traced to different techniques and data used for assessing the attenuation of surface-albedo change by Arctic clouds. For the NH land snow, Chen et al. (2016) estimated that observed changes during 1982–2013 contributed (after converting from NH temperature change to global mean temperature change) by 0.1 W m–2°C–1to global α A, smaller than the estimate of 0.24 W m–2°C–1from Flanner et al. (2011). The contribution of the Southern Hemisphere (SH) to global α A is expected to be small because seasonal snow cover extent in the SH is limited, and trends in SH sea ice extent are relatively flat over much of the satellite record (Section 2.3.2).

CMIP5 and CMIP6 models show moderate spread in global α A, determined from century time scale changes(Qu and Hall, 2014; Schneider et al., 2018; Thackeray and Hall, 2019; Zelinka et al., 2020), owing to variations in modelled sea ice loss and snow cover response in boreal forest regions. The multi-model mean global-scale α A(from all contributions) over the 21st century in CMIP5 models under the RCP8.5 scenario was derived by Schneider et al. (2018) to be 0.40 ± 0.10 W m–2°C–1(one standard deviation). Moreover, they found that modelled α Adoes not decline over the 21st century, despite large losses of snow and sea ice, though a weakened feedback is apparent after 2100. Using the idealized abrupt 4xCO2 , as for the other feedbacks, the estimate of the global-scale albedo feedback in the CMIP5 models is 0.35 ± 0.08 W m–2°C–1(one standard deviation; Vial et al., 2013; Caldwell et al., 2016). The CMIP6 multi-model mean varies from 0.3 to 0.5 W m–2°C–1depending on the kernel used (Zelinka et al., 2020). Donohoe et al. (2020) derived a multi-model mean α Aand its inter-model spread of 0.37 ± 0.19 W m–2°C–1from the CMIP5abrupt 4xCO2 ensemble, employing model-specific estimates of atmospheric attenuation and thereby avoiding bias associated with use of a single radiative kernel.

The surface-albedo feedback estimates using centennial changes have been shown to be highly correlated to those using seasonal regional changes for NH land snow (Qu and Hall, 2014) and Arctic sea ice (Thackeray and Hall, 2019). For the NH land snow, because the physics underpinning this relationship are credible, this opens the possibility to use it as an emergent constraint (Qu and Hall, 2014). Considering only the eight models whose seasonal cycle of albedo feedback falls within the observational range does not change the multi-model mean contribution to global α A(0.08 W m–2°C–1) but decreases the inter-model spread by a factor of two (from ±0.03 to ±0.015 W m–2°C–1; Qu and Hall, 2014). For Arctic sea ice, Thackeray and Hall (2019) show that the seasonal cycle also provides an emergent constraint, at least until mid-century when the relationship degrades. They find that the CMIP5 multi-model mean of the Arctic sea ice contribution to α A is 0.13 W m–2°C–1and that the inter-model spread is reduced by a factor of two (from ±0.04 to ±0.02 W m–2°C–1) when the emergent constraint is used. This model estimate is smaller than observational estimates (Pistone et al., 2014; Cao et al., 2015) except those of Donohoe et al. (2020). This can be traced to CMIP5 models generally underestimating the rate of Arctic sea ice loss during recent decades (Section 9.3.1; Stroeve et al., 2012; Flato et al., 2013), though this may also be an expression of internal variability, since the observed behaviour is captured within large ensemble simulations (Notz, 2015). CMIP6 models better capture the observed Arctic sea ice decline (Section 3.4.1). In the SH the opposite situation is observed. Observations show relatively flat trends in SH sea ice over the satellite era (Section 2.3.2.1) whereas CMIP5 models simulate a small decrease (Section 3.4.1). SH α A is presumably larger in models than observations but only contributes about one quarter of the global α A. Thus, we assess that α Aestimates are consistent, at global scale, in CMIP5 and CMIP6 models and satellite observations, though hemispheric differences and the role of internal variability need to be further explored.

Based on the multiple lines of evidence presented above that include observations, CMIP5 and CMIP6 models and theory, the global surface-albedo feedback is assessed to be positive with high confidence. The basic phenomena that drive this feedback are well understood and the different studies cover a large variety of hypotheses or behaviours, including how the evolution of clouds affects this feedback. The value of the global surface-albedo feedback is assessed to be α A= 0.35 W m–2°C–1, with avery likely range from 0.10 to 0.60 W m–2°C–1and a likely range from 0.25 to 0.45 W m–2°C–1 with high confidence.

7.4.2.4 Cloud Feedbacks

7.4.2.4.1 Decomposition of clouds into regimes

Clouds can be formed almost anywhere in the atmosphere when moist air parcels rise and cool, enabling the water vapour to condense. Clouds consist of liquid water droplets and/or ice crystals, and these droplets and crystals can grow into larger particles of rain, snow or drizzle. These microphysical processes interact with aerosols, radiation and atmospheric circulation, resulting in a highly complex set of processes governing cloud formation and life cycles that operate across a wide range of spatial and temporal scales.

Clouds have various types, from optically thick convective clouds to thin stratus and cirrus clouds, depending upon thermodynamic conditions and large-scale circulation (Figure 7.9). Over the equatorial warm pool and inter-tropical convergence zone (ITCZ) regions, high SSTs stimulate the development of deep convective cloud systems, which are accompanied by anvil and cirrus clouds near the tropopause where the convective air outflows. The large-scale circulation associated with these convective clouds leads to subsidence over the subtropical cool ocean, where deep convection is suppressed by a lower tropospheric inversion layer maintained by the subsidence and promoting the formation of shallow cumulus and stratocumulus clouds. In the extratropics, mid-latitude storm tracks control cloud formation, which occurs primarily in the frontal bands of extratropical cyclones. Since liquid droplets do not freeze spontaneously at temperatures warmer than approximately –40°C and ice nucleating particles that can aid freezing at warmer temperatures are scarce (see (Section 7.3.3), extratropical clouds often consist both of super-cooled liquid and ice crystals, resulting in mixed-phase clouds.

Figure 7.9 | Schematic cross section of diverse cloud responses to surface warming from the tropics to polar regions. Thick solid and dashed curves indicate the tropopause and the subtropical inversion layer in the current climate, respectively. Thin grey text and arrows represent robust responses in the thermodynamic structure to greenhouse warming, of relevance to cloud changes. Text and arrows in red, orange and green show the major cloud responses assessed with high, medium and low confidence, respectively, and the sign of their feedbacks to the surface warming is indicated in the parenthesis. Major advances since AR5 are listed in the box. Figure adapted from Boucher et al. (2013).

In the global energy budget at TOA, clouds affect shortwave (SW) radiation by reflecting sunlight due to their high albedo (cooling the climate system) and also longwave (LW) radiation by absorbing the energy from the surface and emitting at a lower temperature to space, that is, contributing to the greenhouse effect, warming the climate system. In general, the greenhouse effect of clouds strengthens with height whereas the SW reflection depends on the cloud optical properties. The effects of clouds on Earth’s energy budget are measured by the cloud radiative effect (CRE), which is the difference in the TOA radiation between clear and all skies (see (Section 7.2.1). In the present climate, the SW CRE tends to be compensated by the LW CRE over the equatorial warm pool, leading to the net CRE pattern showing large negative values over the eastern part of the subtropical ocean and the extratropical ocean due to the dominant influence of highly reflective marine low-clouds.

In a first attempt to systematically evaluate equilibrium climate sensitivity (ECS) based on fully coupled general circulation models (GCMs) in AR4, diverging cloud feedbacks were recognized as a dominant source of uncertainty. An advance in understanding the cloud feedback was to assess feedbacks separately for different cloud regimes (Gettelman and Sherwood, 2016). A thorough assessment of cloud feedbacks in different cloud regimes was carried out in AR5 (Boucher et al., 2013), which assigned high or medium confidence for some cloud feedbacks butlow or no confidence for others (Table 7.9). Many studies that estimate the net cloud feedback using CMIP5 simulations (Vial et al., 2013; Caldwell et al., 2016; Zelinka et al., 2016; Colman and Hanson, 2017) show different values depending on the methodology and the set of models used, but often report a large inter-model spread of the feedback, with the 90% confidence interval spanning both weak negative and strong positive net feedbacks. Part of this diversity arises from the dependence of the model cloud feedbacks on the parametrization of clouds and their coupling to other sub-grid-scale processes (Zhao et al., 2015).

Since AR5, community efforts have been undertaken to understand and quantify the cloud feedbacks in various cloud regimes coupled with large-scale atmospheric circulation (Bony et al., 2015). For some cloud regimes, alternative tools to ESMs, such as observations, theory, high-resolution cloud resolving models (CRMs), and large eddy simulations (LES), help quantify the feedbacks. Consequently, the net cloud feedback derived from ESMs has been revised by assessing the regional cloud feedbacks separately and summing them with weighting by the ratio of fractional coverage of those clouds over the globe to give the global feedback, following an approach adopted in Sherwood et al. (2020). This ‘bottom-up’ assessment is explained below with a summary of updated confidence of individual cloud feedback components (Table 7.9). Dependence of cloud feedbacks on evolving patterns of surface warming will be discussed in (Section 7.4.4 and is not explicitly taken into account in the assessment presented in this section.

7.4.2.4.2 Assessment for individual cloud regimes

High-cloud altitude feedback

It has long been argued that cloud-top altitude rises under global warming, concurrent with the rising of the tropopause at all latitudes (Marvel et al., 2015; Thompson et al., 2017). This increasing altitude of high-clouds was identified in early generation GCMs and the tropical high-cloud altitude feedback was assessed to be positive with high confidence in AR5 (Boucher et al., 2013). This assessment is supported by a theoretical argument called the ‘fixed anvil temperature mechanism’, which ensures that the temperature of the convective detrainment layer does not change when the altitude of high-cloud tops increases with the rising tropopause (Hartmann and Larson, 2002). Because the cloud-top temperature does not change significantly with global warming, cloud LW emission does not increase even though the surface warms, resulting in an enhancement of the high-cloud greenhouse effect (a positive feedback; Yoshimori et al. (2020)). The upward shift of high-clouds with surface warming is detected in observed interannual variability and trends in satellite records for recent decades (Chepfer et al., 2014; Norris et al., 2016; Saint-Lu et al., 2020). The observational detection is not always successful (Davies et al., 2017), but the cloud altitude shifts similarly in many CRM experiments (Khairoutdinov and Emanuel, 2013; Tsushima et al., 2014; Narenpitak et al., 2017). The high-cloud altitude feedback was estimated to be 0.5 W m–2°C–1based on GCMs in AR5, but is revised, using a recent re-evaluation that excludes aliasing effects by reduced low-cloud amounts, downward to 0.22 ± 0.12 W m–2°C–1(one standard deviation; Zhou et al., 2014; Zelinka et al., 2020). In conclusion, there is high confidence in the positive high-cloud altitude feedback simulated in ESMs as it is supported by theoretical, observational, and process modelling studies.

Tropical high-cloud amount feedback

Updrafts in convective plumes lead to detrainment of moisture at a level where the buoyancy diminishes, and thus deep convective clouds over high SSTs in the tropics are accompanied by anvil and cirrus clouds in the upper troposphere. These clouds, rather than the convective plumes themselves, play a substantial role in the global TOA radiation budget. In the present climate, the net CRE of these clouds is small due to a cancellation between the SW and LW components (Hartmann et al., 2001). However, high-clouds with different optical properties could respond to surface warming differently, potentially perturbing this radiative balance and therefore leading to a non-zero feedback.

A thermodynamic mechanism referred to as the ‘stability iris effect’ has been proposed to explain that the anvil cloud amount decreases with surface warming (Bony et al., 2016). In this mechanism, a temperature-mediated increase of static stability in the upper troposphere, where convective detrainment occurs, acts to balance a weakened mass outflow from convective clouds, and thereby reduce anvil cloud areal coverage (Figure 7.9). The reduction of anvil cloud amount is accompanied by enhanced convective aggregation that causes a drying of the surrounding air and thereby increases the LW emission to space that acts as a negative feedback (Bony et al., 2020). This phenomenon is found in many CRM simulations (Emanuel et al., 2014; Wing and Emanuel, 2014; Wing et al., 2020) and also identified in observed interannual variability (Stein et al., 2017; Saint-Lu et al., 2020).

Despite the reduction of anvil cloud amount supported by several lines of evidence, estimates of radiative feedback due to high-cloud amount changes is highly uncertain in models. The assessment presented here is guided by combined analyses of TOA radiation and cloud fluctuations at interannual time scale using multiple satellite datasets. The observationally based local cloud amount feedback associated with optically thick high-clouds is negative, leading to its global contribution (by multiplying the mean tropical anvil cloud fraction of about 8%) of –0.24 ± 0.05 W m–2°C–1(one standard deviation) for LW (Vaillant de Guélis et al., 2018). Also, there is a positive feedback due to increase of optically thin cirrus clouds in the tropopause layer, estimated to be 0.09 ± 0.09 W m–2°C–1(one standard deviation; Zhou et al., 2014). The negative LW feedback due to reduced amount of thick high-clouds is partly compensated by the positive SW feedback (due to less reflection of solar radiation), so that the tropical high-cloud amount feedback is assessed to be equal to or smaller than their sum. Consistently, the net high-cloud feedback in the tropical convective regime, including a part of the altitude feedback, is estimated to have the global contribution of –0.13 ± 0.06 W m–2°C–1(one standard deviation; Williams and Pierrehumbert, 2017). The negative cloud LW feedback is considerably biased in CMIP5 GCMs (Mauritsen and Stevens, 2015; Su et al., 2017; Li et al., 2019) and highly uncertain, primarily due to differences in the convective parametrization (Webb et al., 2015). Furthermore, high-resolution CRM simulations cannot alone be used to constrain uncertainty because the results depend on parametrized cloud microphysics and turbulence (Bretherton et al., 2014; Ohno et al., 2019). Therefore, the tropical high-cloud amount feedback is assessed as negative but with low confidence given the lack of modelling evidence. Taking observational estimates altogether and methodological uncertainty into account, the global contribution of the high-cloud amount feedback is assessed to be –0.15 ± 0.2 W m–2°C–1(one standard deviation).

Subtropical marine low-cloud feedback

It has long been argued that the response of marine boundary-layer clouds over the subtropical ocean to surface warming was the largest contributor to the spread among GCMs in the net cloud feedback (Boucher et al., 2013). However, uncertainty of the marine low-cloud feedback has been reduced considerably since AR5 through combined knowledge from theoretical, modelling and observational studies (Klein et al., 2017). Processes that control the low-clouds are complex and involve coupling with atmospheric motions on multiple scales, from the boundary-layer turbulence to the large-scale subsidence, which may be represented by a combination of shallow and deep convective mixing (Sherwood et al., 2014).

In order to disentangle the large-scale processes that cause the cloud amount either to increase or decrease in response to the surface warming, the cloud feedback has been expressed in terms of several ‘cloud controlling factors’ (Qu et al., 2014, 2015; Zhai et al., 2015; Brient and Schneider, 2016; Myers and Norris, 2016; McCoy et al., 2017a). The advantage of this approach over conventional calculation of cloud feedbacks is that the temperature-mediated cloud response can be estimated without using information of the simulated cloud responses that are less well-constrained than the changes in the environmental conditions. Two dominant factors are identified for the subtropical low-clouds: a thermodynamic effect due to rising SST that acts to reduce low-cloud by enhancing cloud-top entrainment of dry air, and a stability effect accompanied by an enhanced inversion strength that acts to increase low-cloud (Qu et al., 2014, 2015; Kawai et al., 2017). These controlling factors compensate with a varying degree in different ESMs, but can be constrained by referring to the observed seasonal or interannual relationship between the low-cloud amount and the controlling factors in the environment as a surrogate. The analysis leads to a positive local feedback that has the global contribution of 0.14 to 0.36 W m–2°C–1(Klein et al., 2017), to which the feedback in the stratocumulus regime dominates over the feedback in the trade cumulus regime (Cesana et al., 2019; Radtke et al., 2021). The stratocumulus feedback may be underestimated because explicit simulations using LES show a larger local feedback of up to 2.5 W m–2°C–1, corresponding to the global contribution of 0.2 W m–2°C–1by multiplying the mean tropical stratocumulus fraction of about 8% (Bretherton, 2015). Supported by different lines of evidence, the subtropical marine low-cloud feedback is assessed as positive with high confidence. Based on the combined estimate using LESs and the cloud controlling factor analysis, the global contribution of the feedback due to marine low-clouds equatorward of 30° is assessed to be 0.2 ± 0.16 W m–2°C–1(one standard deviation), for which the range reflects methodological uncertainties.

Land cloud feedback

Intensification of the global hydrological cycle is a robust feature of global warming, but at the same time, many land areas in the subtropics will experience drying at the surface and in the atmosphere (Section 8.2.2). This occurs due to limited water availability in these regions, where the cloudiness is consequently expected to decrease. Reduction in clouds over land is consistently identified in the CMIP5 models and also in a GCM with explicit convection (Bretherton et al., 2014; Kamae et al., 2016a). Because low-clouds make up the majority of subtropical land clouds, this reduced amount of low-clouds reflects less solar radiation and leads to a positive feedback similar to the marine low-clouds. The mean estimate of the global land cloud feedback in CMIP5 models is smaller than the marine low-cloud feedback, 0.08 ± 0.08 W m–2°C–1(Zelinka et al., 2016). These values are nearly unchanged in CMIP6 (Zelinka et al., 2020). However, ESMs still have considerable biases in the climatological temperature and cloud fraction over land, and the magnitude of this feedback has not yet been supported by observational evidence. Therefore, the feedback due to decreasing land clouds is assessed to be 0.08 ± 0.08 W m–2°C–1(one standard deviation) with low confidence.

Mid-latitude cloud amount feedback

Poleward shifts in the mid-latitude jets are evident since the 1980s (Section 2.3.1.4.3) and are a feature of the large-scale circulation change in future projections (Section 4.5.1.6). Because mid-latitude clouds over the North Pacific, North Atlantic and Southern Ocean are induced mainly by extratropical cyclones in the storm tracks along the jets, it has been suggested that the jet shifts should be accompanied by poleward shifts in the mid-latitude clouds, which would result in a positive feedback through the reduced reflection of insolation (Boucher et al., 2013). However, studies since AR5 have revealed that this proposed mechanism does not apply in practice (Ceppi and Hartmann, 2015). While a poleward shift of mid-latitude cloud maxima in the free troposphere has been identified in satellite and ground-based observations (Bender et al., 2012; Eastman and Warren, 2013), associated changes in net CRE are small because the responses in high and low-clouds to the jet shift act to cancel each other (Grise and Medeiros, 2016; Tselioudis et al., 2016; Zelinka et al., 2018). This cancellation is not well captured in ESMs (Lipat et al., 2017), but the above findings show that the mid-latitude cloud feedback is not dynamically driven by the poleward jet shifts, which are rather suggested to occur partly in response to changes in high clouds (Y. Li et al., 2018).

Thermodynamics play an important role in controlling extratropical cloud amount equatorward of about 50° latitude. Recent studies showed, using observed cloud controlling factors, that the mid-latitude low-cloud fractions decrease with rising SST, which also acts to weaken stability of the atmosphere unlike in the subtropics (McCoy et al., 2017a). ESMs consistently show a decrease of cloud amounts and a resultant positive SW feedback in the 30°–40° latitude bands, which can be constrained using observations of seasonal migration of cloud amount (Zhai et al., 2015). Based on the qualitative agreement between observations and ESMs, the mid-latitude cloud amount feedback is assessed as positive with medium confidence. Following these emergent constraint studies using observations and CMIP5/6 models, the global contribution of net cloud amount feedback over 30°–60° ocean areas, covering 27% of the globe, is assessed at 0.09 ± 0.1 W m–2°C–1(one standard deviation), in which the uncertainty reflects potential errors in models’ low-cloud response to changes in thermodynamic conditions.

Extratropical cloud optical depth feedback

Mixed-phase clouds that consist of both liquid and ice are dominant over the Southern Ocean (50°S–80°S), which accounts for 20% of the net CRE in the present climate (Matus and L’Ecuyer, 2017). It has been argued that the cloud optical depth (opacity) will increase over the Southern Ocean as warming drives the replacement of ice-dominated clouds with liquid-dominated clouds (Tan et al., 2019). Liquid clouds generally consist of many small cloud droplets, while the crystals in ice clouds are orders of magnitude fewer in number and much larger, causing the liquid clouds to be optically thicker and thereby resulting in a negative feedback (Boucher et al., 2013). However, this phase-change feedback works effectively only below freezing temperature (Lohmann and Neubauer, 2018; Terai et al., 2019) and other processes that increase or decrease liquid water path (LWP) may also affect the optical depth feedback (McCoy et al., 2019).

Due to insufficient amounts of super-cooled liquid water in the simulated atmospheric mean state, many CMIP5 models overestimated the conversion from ice to liquid clouds with climate warming and the resultant negative phase-change feedback (Kay et al., 2016a; Tan et al., 2016; Lohmann and Neubauer, 2018). This feedback can be constrained using satellite-derived LWP observations over the past 20 years that enable estimates of both long-term trends and the interannual relationship with SST variability (Gordon and Klein, 2014; Ceppi et al., 2016; Manaster et al., 2017). The observationally-constrained SW feedback ranges from –0.91 to –0.46 W m–2°C–1over 40°S–70°S depending on the methodology (Ceppi et al., 2016; Terai et al., 2016). In some CMIP6 models, representation of super-cooled liquid water content has been improved, leading to weaker negative optical depth feedback over the Southern Ocean closer to observational estimates (Bodas-Salcedo et al., 2019; Gettelman et al., 2019). This improvement at the same time results in a positive optical depth feedback over other extratropical ocean where LWP decreased in response to reduced stability in those CMIP6 models (Zelinka et al., 2020). Given the accumulated observational estimates and an improved agreement between ESMs and observations, the extratropical optical depth feedback is assessed to be small negative with medium confidence. Quantitatively, the global contribution of this feedback is assessed to have a value of –0.03 ± 0.05 W m–2°C–1(one standard deviation) by combining estimates based on observed interannual variability and the cloud controlling factors.

Arctic cloud feedback

Clouds in polar regions, especially over the Arctic, form at low altitude above or within a stable to neutral boundary layer and are known to co-vary with sea ice variability beneath. Because the clouds reflect sunlight during summer but trap LW radiation throughout the year, seasonality plays an important role in cloud effects on Arctic climate (Kay et al., 2016b). AR5 assessed that Arctic low-cloud amount will increase in boreal autumn and winter in response to declining sea ice in a warming climate, due primarily to an enhanced upward moisture flux over open water. The cloudier conditions during these seasons result in more downwelling LW radiation, acting as a positive feedback on surface warming (Kay and Gettelman, 2009). Over recent years, further evidence of the cloud contribution to the Arctic amplification has been obtained (Section 7.4.4.1; Goosse et al., 2018). Space-borne lidar (light detection and ranging) observations show that the cloud response to summer sea ice loss is small and cannot overcome the cloud effect in autumn (Taylor et al., 2015; Morrison et al., 2019). The seasonality of the cloud response to sea ice variability is reproduced in GCM simulations (Laîné et al., 2016; Yoshimori et al., 2017). The agreement between observations and models indicates that the Arctic cloud feedback is positive at the surface. This leads to an Arctic cloud feedback at TOA that is likely positive, but very small in magnitude, as found in some climate models (Pithan and Mauritsen, 2014; Morrison et al., 2019). The observational estimates are sensitive to the analysis period and the choice of reanalysis data, and a recent estimate of the TOA cloud feedback over 60°N–90°N using atmospheric reanalysis data and CERES satellite observations suggests a regional value ranging from –0.3 to +0.5 W m–2°C–1, which corresponds to a global contribution of –0.02 to +0.03 W m–2°C–1(R. Zhang et al., 2018). Based on the overall agreement between ESMs and observations, the Arctic cloud feedback is assessed to be small positive and has the value of 0.01 ± 0.05 W m–2°C–1(one standard deviation). The assessed range indicates that a negative feedback is almost as probable as a positive feedback, and the assessment that the Arctic cloud feedback is positive is therefore given low confidence.

7.4.2.4.3 Synthesis for the net cloud feedback

The understanding of the response of clouds to warming and associated radiative feedback has deepened since AR5 (Figure 7.9 and FAQ 7.2). Particular progress has been made in the assessment of the marine low-cloud feedback, which has historically been a major contributor to the cloud feedback uncertainty but is no longer the largest source of uncertainty. Multiple lines of evidence (theory, observations, emergent constraints and process modelling) are now available in addition to ESM simulations, and the positive low-cloud feedback is consequently assessed with high confidence.

The best estimate of net cloud feedback is obtained by summing feedbacks associated with individual cloud regimes and assessed to be α C= 0.42 W m–2°C–1. By assuming that the uncertainties of individual cloud feedbacks are independent of each other, their standard deviations are added in quadrature, leading to the likely range of 0.12 to 0.72 W m–2°C–1and the very likely range of –0.10 to +0.94 W m–2°C–1(Table 7.10). This approach potentially misses feedbacks from cloud regimes that are not assessed, but almost all the major cloud regimes were taken into consideration (Gettelman and Sherwood, 2016) and therefore additional uncertainty will be small. This argument is also supported by an agreement between the net cloud feedback assessed here and the net cloud feedback directly estimated using observations. The observational estimate, which is sensitive to the period considered and is based on two atmospheric reanalyses (ERA-Interim and MERRA) and TOA radiation budgets derived from the CERES satellite observations for the years 2000–2010, is 0.54 ± 0.7 W m–2°C–1(one standard deviation; Dessler, 2013). The observational estimate overlaps with the assessed range of the net cloud feedback. The assessed very likely range is reduced by about 50% compared to AR5, but is still wide compared to those of other climate feedbacks (Table 7.10). The largest contribution to this uncertainty range is the estimate of tropical high-cloud amount feedback which is not yet well quantified using models.

In reality, different types of cloud feedback may occur simultaneously in one cloud regime. For example, an upward shift of high-clouds associated with the altitude feedback could be coupled to an increase/decrease of cirrus/anvil cloud fractions associated with the cloud amount feedback. Alternatively, slowdown of the tropical circulation with surface warming (Section 4.5.3 and Figure 7.9) could affect both high and low-clouds so that their feedbacks are co-dependent. Quantitative assessments of such covariances require further knowledge about cloud feedback mechanisms, which will further narrow the uncertainty range.

In summary, deepened understanding of feedback processes in individual cloud regimes since AR5 leads to an assessment of the positive net cloud feedback with high confidence. A small probability (less than 10%) of a net negative cloud feedback cannot be ruled out, but this would require an extremely large negative feedback due to decreases in the amount of tropical anvil clouds or increases in optical depth of extratropical clouds over the Southern Ocean; neither is supported by current evidence.

Table 7.9 | Assessed sign and confidence level of cloud feedbacks in different regimes in AR5 and AR6. For some cloud regimes, the feedback was not assessed in AR5, indicated by N/A.

Feedback

AR5

AR6

High-cloud altitude feedback

Positive (high confidence)

Positive (high confidence)

Tropical high-cloud amount feedback

N/A

Negative (low confidence)

Subtropical marine low-cloud feedback

N/A (low confidence)

Positive (high confidence)

Land cloud feedback

N/A

Positive (low confidence)

Mid-latitude cloud amount feedback

Positive (medium confidence)

Positive (medium confidence)

Extratropical cloud optical depth feedback

N/A

Small negative (medium confidence)

Arctic cloud feedback

Small positive (very low confidence)

Small positive (low confidence)

Net cloud feedback

Positive (medium confidence)

Positive (high confidence)

7.4.2.5 Biogeophysical and Non-CO2 Biogeochemical Feedbacks

The feedbacks presented in the previous sections (Sections 7.4.2.1–7.4.2.4) are directly linked to physical climate variables (for example temperature, water vapour, clouds, or sea ice). The central role of climate feedbacks associated with these variables has been recognized since early studies of climate change. However, in addition to these physical climate feedbacks, the Earth system includes feedbacks for which the effect of global mean surface temperature change on the TOA energy budget is mediated through other mechanisms, such as the chemical composition of the atmosphere, or by vegetation changes. Among these additional feedbacks, the most important is the CO2 feedback that describes how a change of the global surface temperature affects the atmospheric CO2 concentration. In ESM simulations in which CO2 emissions are prescribed, changes in surface carbon fluxes affect the CO2 concentration in the atmosphere, the TOA radiative energy budget, and eventually the global mean surface temperature. In ESM simulations in which the CO2 concentration is prescribed, changes in the carbon cycle allow compatible CO2 emissions to be calculated, that is, the CO2 emissions that are compatible with both the prescribed CO2 concentration and the representation of the carbon cycle in the ESM. The CO2 feedback is assessed in (Chapter 5 Section 5.4). The framework presented in this chapter assumes that the CO2 concentration is prescribed, and our assessment of the net feedback parameter, α , does not include carbon cycle feedbacks on the atmospheric CO2 concentration (Section 7.1 and Box 7.1). However, our assessment of α does include non-CO2 biogeochemical feedbacks (including effects due to changes in atmospheric methane concentration; Section 7.4.2.5.1) and biogeophysical feedbacks (Section 7.4.2.5.2). A synthesis of the combination of biogeophysical and non-CO2 biogeochemical feedbacks is given in Section 7.4.2.5.3.

7.4.2.5.1 Non-CO2biogeochemical feedbacks

The chemical composition of the atmosphere (beyond CO2 and water vapour changes) is expected to change in response to a warming climate. These changes in greenhouse gases (methane, nitrous oxide and ozone) and aerosol amount (including dust) have the potential to alter the TOA energy budget and are collectively referred to as ‘non-CO2 biogeochemical feedbacks’. Methane (CH4) and nitrous oxide (N2O) feedbacks arise partly from changes in their emissions from natural sources in response to temperature change; these are assessed in (Chapter 5 Section 5.4.7; see also Figure 5.29c). Here we exclude the permafrost CH4 feedback (Section 5.4.9.1.2) because, although associated emissions are projected to increase under warming on multi-decadal to centennial time scales, on longer time scales these emissions would eventually substantially decline as the permafrost carbon pools were depleted (Schneider von Deimling et al., 2012, 2015). This leaves the wetland CH4, land N2O, and ocean N2O feedbacks, the assessed mean values of which sum to a positive feedback parameter of +0.04 [0.02 to 0.06] W m–2°C–1Section 5.4.7. Other non-CO2 biogeochemical feedbacks that are relevant to the net feedback parameter are assessed in Chapter 6 (Section 6.4.5 and Table 6.8). These feedbacks are associated with sea salt, dimethyl sulphide, dust, ozone, biogenic volatile organic compounds, lightning, and CH4 lifetime, and sum to a negative feedback parameter of –0.20 [–0.41 to +0.01] W m–2°C–1. The overall feedback parameter for non-CO2 biogeochemical feedbacks is obtained by summing the Chapter 5 and Chapter 6 assessments, which gives –0.16 [–0.37 to +0.05] W m–2°C–1. However, there is low confidence in the estimates of both the individual non-CO2 biogeochemical feedbacks as well as their total effect, as evident from the large range in the magnitudes of α from different studies, which can be attributed to diversity in how models account for these feedbacks and limited process-level understanding.

7.4.2.5.2 Biogeophysical feedbacks

Biogeophysical feedbacks are associated with changes in the spatial distribution and/or biophysical properties of vegetation, induced by surface temperature change and attendant hydrological cycle change. These vegetation changes can alter radiative fluxes directly via albedo changes, or via surface momentum or moisture flux changes and hence changes in cloud properties. However, the direct physiological response of vegetation to changes in CO2, including changes in stomatal conductance, is considered part of the CO2 effective radiative forcing rather than a feedback (Section 7.3.2.1). The time scale on which vegetation responds to climate change is relatively uncertain but can be from decades to hundreds of years (Willeit et al., 2014), and could occur abruptly or as a tipping point (Sections 5.4.9.1.1, 8.6.2.1 and 8.6.2.2); equilibrium only occurs when the soil system and associated nutrient and carbon pools equilibrate, which can take millennia (Brantley, 2008; Sitch et al., 2008). The overall effects of climate-induced vegetation changes may be comparable in magnitude to those from anthropogenic land-use and land-cover change (Davies-Barnard et al., 2015). Climate models that include a dynamical representation of vegetation (e.g., Reick et al., 2013; Harper et al., 2018) are used to explore the importance of biogeophysical feedbacks (Notaro et al., 2007; Brovkin et al., 2009; O’ishi et al., 2009; Port et al., 2012; Willeit et al., 2014; Alo and Anagnostou, 2017; W. Zhang et al., 2018; Armstrong et al., 2019). In AR5, it was discussed that such model experiments predicted that expansion of vegetation in the high latitudes of the Northern Hemisphere would enhance warming due to the associated surface-albedo change, and that reduction of tropical forests in response to climate change would lead to regional surface warming, due to reduced evapotranspiration (M. Collins et al., 2013), but there was no assessment of the associated feedback parameter. The SRCCL stated that regional climate change can be dampened or enhanced by changes in local land cover, but that this depends on the location and the season; however, in general the focus was on anthropogenic land-cover change, and no assessment of the biogeophysical feedback parameter was carried out. There are also indications of a marine biogeophysical feedback associated with surface-albedo change due to changes in phytoplankton (Frouin and Iacobellis, 2002; Park et al., 2015), but there is not currently enough evidence to quantitatively assess this feedback.

Since AR5, several studies have confirmed that a shift from tundra to boreal forests and the associated albedo change leads to increased warming in Northern Hemisphere high latitudes (high confidence) (Willeit et al., 2014; W. Zhang et al., 2018; Armstrong et al., 2019). However, regional modelling indicates that vegetation feedbacks may act to cool climate in the Mediterranean (Alo and Anagnostou, 2017), and in the tropics and subtropics the regional response is in general not consistent across models. On a global scale, several modelling studies have either carried out a feedback analysis (Stocker et al., 2013; Willeit et al., 2014) or presented simulations that allow a feedback parameter to be estimated (O’ishi et al., 2009; Armstrong et al., 2019), in such a way that the physiological response can be accounted for as a forcing rather than a feedback. The central estimates of the biogeophysical feedback parameter from these studies range from close to zero (Willeit et al., 2014) to +0.13 W m–2°C–1(Stocker et al., 2013). An additional line of evidence comes from the mid-Pliocene warm period (MPWP, Chapter 2, Cross-Chapter Box 2.1), for which paleoclimate proxies provide evidence of vegetation distribution and CO2 concentrations. Model simulations that include various combinations of modern versus MPWP vegetation and CO2 allow an associated feedback parameter to be estimated, as long as account is also taken of the orographic forcing (Lunt et al., 2010, 2012b). This approach has the advantage over pure modelling studies in that the reconstructed vegetation is based on (paleoclimate) observations, and is in equilibrium with the CO2 forcing. However, there are uncertainties in the vegetation reconstruction in regions with little or no proxy data, and it is uncertain how much of the vegetation change is associated with the physiological response to CO2. This paleoclimate approach gives an estimate for the biogeophysical feedback parameter of +0.3 W m–2°C–1.

Given the limited number of studies, we take the full range of estimates discussed above for the biogeophysical feedback parameter, and assess the very likely range to be from 0.0 to +0.3 W m–2°C–1, with a central estimate of +0.15 W m–2°C–1(low confidence). Although this assessment is based on evidence from both models and paleoclimate proxies, and the studies above agree on the sign of the change, there is nonetheless limited evidence. Higher confidence could be obtained if there were more studies that allowed calculation of a biogeophysical feedback parameter (particularly from paleoclimates), and if the partitioning between biogeophysical feedbacks and physiological forcing were clearer for all lines of evidence.

7.4.2.5.3 Synthesis of biogeophysical and non-CO2 biogeochemical feedbacks

The non-CO2 biogeochemical feedbacks are assessed in (Section 7.4.2.5.1 to be –0.16 [–0.37 to +0.05] W m2°C–1and the biogeophysical feedbacks are assessed in (Section 7.4.2.5.2 to be +0.15 [0.0 to +0.3] W m–2°C–1. The sum of the biogeophysical and non-CO2 biogeochemical feedbacks is assessed to have a central value of –0.01 W m–2°C–1and avery likely range from –0.27 to +0.25 W m–2°C–1(Table 7.10). Given the relatively long time scales associated with the biological processes that mediate the biogeophysical and many of the non-CO2 biogeochemical feedbacks, in comparison with the relatively short time scale of many of the underlying model simulations, combined with the small number of studies for some of the feedbacks, and the relatively small signals, this overall assessment has low confidence.

Some supporting evidence for this overall assessment can be obtained from the CMIP6 ensemble, which provides some pairs of instantaneous 4×CO2 simulations carried out using related models, with and without biogeophysical and non-CO2 biogeochemical feedbacks. This is not a direct comparison because these pairs of simulations may differ by more than just their inclusion of these additional feedbacks; furthermore, not all biogeophysical and non-CO2 biogeochemical feedbacks are fully represented. However, a comparison of the pairs of simulations does provide a first-order estimate of the magnitude of these additional feedbacks. Séférian et al. (2019) find a slightly more negative feedback parameter in CNRM-ESM2-1 (with additional feedbacks) then in CNRM-CM6-1 (a decrease of 0.02 W m–2°C–1, using the linear regression method from years 10–150). Andrews et al. (2019) also find a slightly more negative feedback parameter when these additional feedbacks are included (a decrease of 0.04 W m–2°C–1in UKESM1 compared with HadGEM3-GC3.1). Both of these studies suggest a small but slightly negative feedback parameter for the combination of biogeophysical and non-CO2 biogeochemical feedbacks, but with relatively large uncertainty given (i) interannual variability and (ii) that feedbacks associated with natural terrestrial emissions of CH4 and N2O were not represented in either pair.

7.4.2.6 Long-Term Radiative Feedbacks Associated with Ice Sheets

Although long-term radiative feedbacks associated with ice sheets are not included in our definition of ECS (Box 7.1), the relevant feedback parameter is assessed here because the time scales on which these feedbacks act are relatively uncertain, and the long-term temperature response to CO2 forcing of the entire Earth system may be of interest.

Earth’s ice sheets (Greenland and Antarctica) are sensitive to climate change (Section 9.4; Pattyn et al., 2018). Their time evolution is determined by both their surface mass balance and ice dynamic processes, with the latter being particularly important for the West Antarctic Ice Sheet. Surface mass balance depends on the net energy and hydrological fluxes at their surface, and there are mechanisms of ice-sheet instability that depend on ocean temperatures and basal melt rates (Section 9.4.1.1). The presence of ice sheets affects Earth’s radiative budget, hydrology, and atmospheric circulation due to their characteristic high albedo, low roughness length, and high altitude, and they influence ocean circulation through freshwater input from calving and melt (e.g., Fyke et al., 2018). Ice-sheet changes also modify surface albedo through the attendant change in sea level and therefore land area (Abe-Ouchi et al., 2015). The time scale for ice sheets to reach equilibrium is of the order of thousands of years (Clark et al., 2016). Due to the long time scales involved, it is a major challenge to run coupled climate–ice sheet models to equilibrium, and as a result, long-term simulations are often carried out with lower complexity models, and/or are asynchronously coupled.

In AR5, it was described that both the Greenland and Antarctic ice sheets would continue to lose mass in a warming world (M. Collins et al., 2013), with a continuation in sea level rise beyond the year 2500 assessed as virtually certain. However, there was low confidence in the associated radiative feedback mechanisms, and as such, there was no assessment of the magnitude of long-term radiative feedbacks associated with ice sheets. That assessment is consistent with SROCC, wherein it was stated that ‘with limited published studies to draw from and no simulations run beyond 2100, firm conclusions regarding the net importance of atmospheric versus ocean melt feedbacks on the long-term future of Antarctica cannot be made.’

The magnitude of the radiative feedback associated with changes to ice sheets can be quantified by comparing the global mean long-term equilibrium temperature response to increased CO2 concentrations in simulations that include interactive ice sheets with that of simulations that do not include the associated ice sheet–climate interactions (Swingedouw et al., 2008; Vizcaíno et al., 2010; Goelzer et al., 2011; Bronselaer et al., 2018; Golledge et al., 2019). These simulations indicate that on multi-centennial time scales, ice-sheet mass loss leads to freshwater fluxes that can modify ocean circulation (Swingedouw et al., 2008; Goelzer et al., 2011; Bronselaer et al., 2018; Golledge et al., 2019). This leads to reduced surface warming (by about 0.2°C in the global mean after 1000 years; Section 7.4.4.1.1; Goelzer et al., 2011), although other work suggests no net global temperature effect of ice-sheet mass loss (Vizcaíno et al., 2010). However, model simulations in which the Antarctic Ice Sheet is removed completely in a paleoclimate context indicate a positive global mean feedback on multi-millennial time scales due primarily to the surface-albedo change (Goldner et al., 2014a; Kennedy-Asser et al., 2019); in (Chapter 9 Section 9.6.3) it is assessed that such ice-free conditions could eventually occur given 7°C–13°C of warming. This net positive feedback from ice-sheet mass loss on long time scales is also supported by model simulations of the mid-Pliocene Warm Period (MPWP; Cross-chapter Box 2.1) in which the volume and area of the Greenland and West Antarctic ice sheets are reduced in model simulations in agreement with geological data (Chandan and Peltier, 2018), leading to surface warming. As such, overall, on multi-centennial time scales the feedback parameter associated with ice sheets is likely negative (medium confidence), but on multi-millennial time scales by the time the ice sheets reach equilibrium, the feedback parameter is very likely positive (high confidence) (Table 7.10). However, a relative lack of models carrying out simulations with and without interactive ice sheets over centennial to millennial time scales means that there is currently not enough evidence to quantify the magnitude of these feedbacks, or the time scales on which they act.

7.4.2.7 Synthesis

Table 7.10 summarizes the estimates and the assessment of the individual and the net feedbacks presented in the above sections. The uncertainty range of the net climate feedback was obtained by adding standard deviations of individual feedbacks in quadrature, assuming that they are independent and follow the Gaussian distribution. It is virtually certain that the net climate feedback is negative, primarily due to the Planck temperature response, indicating that climate acts to stabilize in response to radiative forcing imposed to the system. Supported by the level of confidence associated with the individual feedbacks, it is also virtually certain that the sum of the non-Planck feedbacks is positive. Based on Table 7.10 these climate feedbacks amplify the Planck temperature response by about 2.8 [1.9 to 5.9] times . Cloud feedback remains the largest contributor to uncertainty of the net feedback, but the uncertainty is reduced compared to AR5. A secondary contribution to the net feedback uncertainty is the biogeophysical and non-CO2 biogeochemical feedbacks, which together are assessed to have a central value near zero and thus do not affect the central estimate of ECS. The net climate feedback is assessed to be –1.16 W m–2°C–1, likely from –1.54 to –0.78 W m–2°C–1, and very likely from –1.81 to –0.51 W m–2°C–1.

Feedback parameters in climate models are calculated assuming that they are independent of each other, except for a well-known co-dependency between the water vapour (WV) and lapse rate (LR) feedbacks. When the inter-model spread of the net climate feedback is computed by adding in quadrature the inter-model spread of individual feedbacks, it is 17% wider than the spread of the net climate feedback directly derived from the ensemble. This indicates that the feedbacks in climate models are partly co-dependent. Two possible co-dependencies have been suggested (Huybers, 2010; Caldwell et al., 2016). One is a negative covariance between the LR and longwave cloud feedbacks, which may be accompanied by a deepening of the troposphere (O’Gorman and Singh, 2013; Yoshimori et al., 2020) leading both to greater rising of high-clouds and a larger upper-tropospheric warming. The other is a negative covariance between albedo and shortwave cloud feedbacks, which may originate from the Arctic regions: a reduction in sea ice enhances the shortwave cloud radiative effect because the ocean surface is darker than sea ice (Gilgen et al., 2018). This covariance is reinforced as the decrease of sea ice leads to an increase in low-level clouds (Mauritsen et al., 2013). However, the mechanism causing these co-dependences between feedbacks is not well understood yet and a quantitative assessment based on multiple lines of evidence is difficult. Therefore, this synthesis assessment does not consider any co-dependency across individual feedbacks.

The assessment of the net climate feedback presented above is based on a single approach (i.e., process understanding) and directly results in a value for ECS given in (Section 7.5.1; this is in contrast to the synthesis assessment of ECS in (Section 7.5.5 which combines multiple approaches. The total (net) feedback parameter consistent with the final synthesis assessment of the ECS and Equation 7.1 (Box 7.1) is provided there.

7.4.2.8 Climate Feedbacks in ESMs

Since AR5, many modelling groups have newly participated in CMIP experiments, leading to an increase in the number of models in CMIP6 Section 1.5.4). Other modelling groups that contributed to CMIP5 also updated their ESMs for carrying out CMIP6 experiments. While some of the CMIP6 models share components and are therefore not independent, they are analysed independently when calculating climate feedbacks. This, and more subtle forms of model inter-dependence, creates challenges when determining appropriate model weighting schemes (Section 1.5.4). Additionally, it must be kept in mind that the ensemble sizes of the CMIP5 and CMIP6 models are not sufficiently large to sample the full range of model uncertainty.

The multi-model mean values of all physical climate feedbacks are calculated using the radiative kernel method (Section 7.4.1) and compared with the assessment in the previous sections (Figure 7.10). For CMIP models, there is a discrepancy between the net climate feedback calculated directly using the time evolutions of ΔT and ΔN in each model and the accumulation of individual feedbacks, but it is negligibly small (Supplementary Material 7.SM.4). Feedbacks due to biogeophysical and non-CO2 biogeochemical processes are included in some models but neglected in the kernel analysis. In AR6, biogeophysical and non-CO2 biogeochemical feedbacks are explicitly assessed (Section 7.4.2.5).

Table 7.10 | Synthesis assessment of climate feedbacks (central estimate shown in bold). The mean values and their 90% ranges in CMIP5/6 models, derived using multiple radiative kernels (Zelinka et al., 2020) are also presented for comparison.

Feedback Parameter α x(W m–2°C–1)

CMIP5 GCMs

CMIP6 ESMs

AR6 Assessed Ranges

Mean and

5–95% Interval

Mean and

5–95% Interval

Central Estimate

Very likely Interval

Likely Interval

Level of Confidence

Planck

–3.20 [–3.3 to –3.1]

–3.22 [–3.3 to –3.1]

–3.22

–3.4 to –3.0

–3.3 to –3.1

high

WV+LR

1.24 [1.08 to 1.35]

1.25 [1.14 to 1.45]

1.30

1.1 to 1.5

1.2 to 1.4

high

Surface albedo

0.41 [0.25 to 0.56]

0.39 [0.26 to 0.53]

0.35

0.10 to 0.60

0.25 to 0.45

medium

Clouds

0.41 [–0.09 to 1.1]

0.49 [–0.08 to 1.1]

0.42

–0.10 to 0.94

0.12 to 0.72

high

Biogeophysical and non-CO2 biogeochemical

Not evaluated

Not evaluated

–0.01

–0.27 to 0.25

–0.16 to 0.14

low

Residual of kernel estimates

0.06 [–0.17 to 0.29]

0.05 [–0.18 to 0.28 ]

Net (i.e., relevant for ECS)

–1.08 [–1.61 to –0.68]

–1.03 [–1.54 to –0.62]

–1.16

–1.81 to –0.51

–1.54 to –0.78

medium

Long-term ice-sheet feedbacks (millennial scale)

>0.0

high

Figure 7.10 | Global mean climate feedbacks estimated in abrupt 4xCO2 simulations of 29 CMIP5 models (light blue) and 49 CMIP6 models (orange), compared with those assessed in this Report (red). Individual feedbacks for CMIP models are averaged across six radiative kernels as computed in Zelinka et al. (2020). The white line, black box and vertical line indicate the mean, 66% and 90% ranges, respectively. The shading represents the probability distribution across the full range of GCM/ESM values and for the 2.5–97.5 percentile range of the AR6 normal distribution. The unit is W m–2°C–1. Feedbacks associated with biogeophysical and non-CO2 biogeochemical processes are assessed in AR6, but they are not explicitly estimated from general circulation models (GCMs)/Earth system models (ESMs) in CMIP5 and CMIP6. Further details on data sources and processing are available in the chapter data table (Table 7.SM.14).

All the physical climate feedbacks apart from clouds are very similar in the CMIP5 and CMIP6 model ensembles (see also Table 7.10). These values, where possible supported by other lines of evidence, are used for assessing feedbacks in Sections 7.4.2.1–7.4.2.3. A difference found between CMIP5 and CMIP6 models is the net cloud feedback, which is larger in CMIP6 by about 20%. This change is the major cause of less-negative values of the net climate feedback in CMIP6 than in CMIP5 and hence an increase in modelled ECs (Section 7.5.1).

A remarkable improvement of cloud representation in some CMIP6 models is the reduced error of the too-weak negative shortwave CRE over the Southern Ocean (Bodas-Salcedo et al., 2019; Gettelman et al., 2019) due to a more realistic simulation of supercooled liquid droplets and associated cloud optical depths that were biased low commonly in CMIP5 models (McCoy et al., 2014a, b). Because the negative cloud optical depth feedback occurs due to ‘brightening’ of clouds via phase change from ice to liquid cloud particles in response to surface warming (Cesana and Storelvmo, 2017), the extratropical cloud shortwave feedback tends to be less negative or even slightly positive in models with reduced errors (Bjordal et al., 2020; Zelinka et al., 2020). The assessment of cloud feedbacks in (Section 7.4.2.4 incorporates estimates from these improved ESMs. Yet, there still remain other shared model errors, such as in the subtropical low-clouds (Calisto et al., 2014) and tropical anvil clouds (Mauritsen and Stevens, 2015), hampering an assessment of feedbacks associated with these cloud regimes based only on ESMs (Section 7.4.2.4).

7.4.3 Dependence of Feedbacks on Climate Mean State

In the standard framework of forcings and feedbacks (Section 7.4.1 and Box 7.1), the approximation is made that the strength of climate feedbacks is independent of the background global mean surface temperature. More generally, the individual feedback parameters, α x, are often assumed to be constant over a range of climate states, including those reconstructed from the past (encompassing a range of states warmer and colder than today, with varying continental geographies) or projected for the future. If this approximation holds, then the equilibrium global surface temperature response to a fixed radiative forcing will be constant, regardless of the climate state to which that forcing is applied.

This approximation will break down if climate feedbacks are not constant, but instead vary as a function of, for example, background temperature (Roe and Baker, 2007; Zaliapin and Ghil, 2010; Roe and Armour, 2011; Bloch-Johnson et al., 2015), continental configuration (Farnsworth et al., 2019), or configuration of ice sheets (Yoshimori et al., 2009). If the real climate system exhibits this state-dependence, then the future equilibrium temperature change in response to large forcing may be different from that inferred using the standard framework, and/or different to that inferred from paleoclimates. Such considerations are important for the assessment of ECs (Section 7.5). Climate models generally include representations of feedbacks that allow state-dependent behaviour, and so model results may also differ from the predictions from the standard framework.

In AR5 (Boucher et al., 2013), there was a recognition that climate feedbacks could be state-dependent (Colman and McAvaney, 2009), but modelling studies that explored this (e.g., Manabe and Bryan, 1985; Voss and Mikolajewicz, 2001; Stouffer and Manabe, 2003; Hansen et al., 2005b) were not assessed in detail. Also in AR5 (Masson-Delmotte et al., 2013), it was assessed that some models exhibited weaker sensitivity to Last Glacial Maximum (LGM; Cross-Chapter Box 2.1) forcing than to 4×CO2 forcing, due to state-dependence in shortwave cloud feedbacks.

Here, recent evidence for state-dependence in feedbacks from modelling studies (Section 7.4.3.1) and from the paleoclimate record (Section 7.4.3.2) are assessed, with an overall assessment in (Section 7.4.3.3. The focus is on temperature-dependence of feedbacks when the system is in equilibrium with the forcing; evidence for transient changes in the net feedback parameter associated with evolving spatial patterns of warming is assessed separately in (Section 7.4.4.

7.4.3.1 State-dependence of Feedbacks in Models

There are several modelling studies since AR5 in which ESMs of varying complexity have been used to explore temperature dependence of feedbacks, either under modern (Hansen et al., 2013; Jonko et al., 2013; Meraner et al., 2013; Good et al., 2015; Duan et al., 2019; Mauritsen et al., 2019; Rohrschneider et al., 2019; Stolpe et al., 2019; Bloch-Johnson et al., 2020; Rugenstein et al., 2020) or paleo (Caballero and Huber, 2013; Zhu et al., 2019a) climate conditions, typically by carrying out multiple simulations across successive CO2 doublings. A non-linear temperature response to these successive doublings may be partly due to forcing that increases more (or less) than expected from a purely logarithmic dependence (Section 7.3.2; Etminan et al., 2016), and partly due to state-dependence in feedbacks; however, not all modelling studies have partitioned the non-linearities in temperature response between these two effects. Nonetheless, there is general agreement among ESMs that the net feedback parameter, α , increases (i.e., becomes less negative) as temperature increases from pre-industrial levels (i.e., sensitivity to forcing increases as temperature increases; e.g., Meraner et al., 2013; see Figure 7.11). The associated increase in sensitivity to forcing is, in most models, due to the water vapour (Section 7.4.2.2) and cloud (Section 7.4.2.4) feedback parameters increasing with warming (Caballero and Huber, 2013; Meraner et al., 2013; Zhu et al., 2019a; Rugenstein et al., 2020; Sherwood et al., 2020). These changes are offset partially by the surface-albedo feedback parameter decreasing (Jonko et al., 2013; Meraner et al., 2013; Rugenstein et al., 2020), as a consequence of a reduced amount of snow and sea ice cover in a much warmer climate. At the same time, there is little change in the Planck response (Section 7.4.2.1), which has been shown in one model to be due to competing effects from increasing Planck emission at warmer temperatures and decreasing planetary emissivity due to increased CO2 and water vapour (Mauritsen et al., 2019). Analysis of the spatial patterns of the non-linearities in temperature response (Good et al., 2015) suggests that these patterns are linked to a reduced weake