Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Table of Content 2 3 Executive Summary ................................................................................................................................... 4 4 4.1 Scope and Overview of this Chapter ................................................................................................. 9 5 4.2 Methodology ..................................................................................................................................... 11 6 4.2.1 Models, Model Intercomparison Projects, and Ensemble Methodologies ...................................... 11 7 4.2.2 Scenarios ..................................................................................................................................... 13 8 4.2.3 Sources of Near-Term Information............................................................................................... 14 9 4.2.4 Pattern Scaling ............................................................................................................................. 17 10 4.2.5 Quantifying Various Sources of Uncertainty ................................................................................ 18 11 4.2.6 Display of Model Agreement and Spread ..................................................................................... 20 12 13 BOX 4.1: Ensemble Evaluation and Weighting ................................................................................. 21 14 15 4.3 Projected Changes in Global Climate Indices in the 21st Century ................................................. 24 16 4.3.1 Atmosphere ................................................................................................................................. 25 17 4.3.1.1 Surface Air Temperature .............................................................................................................. 25 18 4.3.1.2 Precipitation ................................................................................................................................ 27 19 4.3.2 Cryosphere, Ocean, and Biosphere ............................................................................................... 28 20 4.3.2.1 Arctic Sea Ice .............................................................................................................................. 28 21 4.3.2.2 Global Mean Sea Level ................................................................................................................ 30 22 4.3.2.3 Atlantic Meridional Overturning Circulation ................................................................................ 30 23 4.3.2.4 Ocean and Land Carbon Uptake ................................................................................................... 31 24 4.3.2.5 Surface Ocean pH ........................................................................................................................ 32 25 4.3.3 Modes of Variability .................................................................................................................... 32 26 4.3.3.1 Northern and Southern Annular Modes ........................................................................................ 32 27 4.3.3.2 El Niño-Southern Oscillation ....................................................................................................... 34 28 4.3.4 Synthesis Assessment of Projected Change in Global Surface Air Temperature ............................ 35 29 30 4.4 Near-term Global Climate Changes ................................................................................................ 39 31 4.4.1 Atmosphere ................................................................................................................................. 39 32 4.4.1.1 Average Global Surface Air Temperature..................................................................................... 39 33 4.4.1.2 Spatial Patterns of Surface Warming ............................................................................................ 40 34 4.4.1.3 Precipitation ................................................................................................................................ 40 35 4.4.1.4 Global Monsoon Precipitation and Circulation ............................................................................. 41 36 4.4.2 Cryosphere, Ocean, and Biosphere ............................................................................................... 42 37 4.4.2.1 Arctic Sea Ice .............................................................................................................................. 42 38 4.4.2.2 Ocean and Land Carbon flux ........................................................................................................ 43 39 4.4.3 Modes of Variability .................................................................................................................... 44 40 4.4.3.1 Northern and Southern Annular Modes ........................................................................................ 44 41 4.4.3.2 El Niño-Southern Oscillation ....................................................................................................... 45 42 4.4.3.3 Indian Ocean Basin and Dipole Modes. ........................................................................................ 46 43 4.4.3.4 Tropical Atlantic Modes .............................................................................................................. 46 44 4.4.3.5 Pacific Decadal Variability .......................................................................................................... 47 45 4.4.3.6 Atlantic Multidecadal Variability ................................................................................................. 47 46 4.4.4 Response to Short-Lived Climate Forcers and Volcanic Eruptions ............................................... 48 47 48 Cross-Chapter Box 4.1: The climate effects of volcanic eruption........................................................ 51 49 50 4.5 Mid- to Long-term Global Climate Change .................................................................................... 53 51 4.5.1 Atmosphere ................................................................................................................................. 53 52 4.5.1.1 Near-Surface Air Temperature ..................................................................................................... 53 Do Not Cite, Quote or Distribute 4-2 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 4.5.1.2 Annual Mean Atmospheric Temperature ...................................................................................... 57 2 4.5.1.3 Near-Surface Relative Humidity .................................................................................................. 58 3 4.5.1.4 Precipitation ................................................................................................................................ 60 4 4.5.1.5 Global Monsoon Precipitation and Circulation ............................................................................. 61 5 4.5.1.6 Sea Level Pressure, Large-scale Atmospheric Circulation, Storm Tracks and Blocking ................ 62 6 4.5.2 Ocean .......................................................................................................................................... 67 7 4.5.2.1 Ocean Temperature ...................................................................................................................... 67 8 4.5.2.2 Ocean acidification ...................................................................................................................... 67 9 4.5.3 Modes of Variability .................................................................................................................... 68 10 4.5.3.1 Northern and Southern Annular Modes ........................................................................................ 68 11 4.5.3.2 El Niño-Southern Oscillation ....................................................................................................... 69 12 4.5.3.3 Indian Ocean Basin and Dipole Modes ......................................................................................... 70 13 4.5.3.4 Tropical Atlantic Modes .............................................................................................................. 71 14 4.5.3.5 Pacific Decadal Variability .......................................................................................................... 71 15 4.5.3.6 Atlantic Multidecadal Variability ................................................................................................. 71 16 17 4.6 Implications of Climate Policy ......................................................................................................... 72 18 4.6.1 Patterns of Climate Change for Specific Levels of Global ............................................................ 72 19 4.6.1.1 Temperature ................................................................................................................................ 73 20 4.6.1.2 Precipitation ................................................................................................................................ 74 21 4.6.1.3 Atmospheric Circulation .............................................................................................................. 75 22 4.6.2 Climate Goals, Overshoot, and Path-Dependence ......................................................................... 76 23 4.6.2.1 Climate change under overshoot .................................................................................................. 76 24 4.6.2.2 Consistency between Shared Socioeconomic Pathways and Representative Concentration Pathways 25 .................................................................................................................................................... 77 26 4.6.3 Climate Response to mitigation, Carbon Dioxide Removal, and Solar Radiation Modification ..... 78 27 4.6.3.1 Emergence of the climate response to mitigation .......................................................................... 79 28 4.6.3.2 Climate Response to Mitigation by Carbon Dioxide Removal ...................................................... 81 29 4.6.3.3 Climate Response to Solar Radiation Modification....................................................................... 83 30 31 4.7 Climate Change Beyond 2100 .......................................................................................................... 91 32 4.7.1 Commitment and Climate Change Beyond 2100 .......................................................................... 91 33 4.7.1.1 Climate change following zero emissions ..................................................................................... 91 34 4.7.1.2 Change in Global Climate Indices Beyond 2100........................................................................... 93 35 4.7.1.2.1 Global Surface Air Temperature.................................................................................................. 94 36 4.7.1.2.2 Global Land Precipitation ........................................................................................................... 94 37 4.7.1.2.3 Arctic Sea Ice ............................................................................................................................. 95 38 4.7.2 Potential for Abrupt and Irreversible Climate Change................................................................... 95 39 40 4.8 Low-Likelihood High-Warming Storylines ..................................................................................... 97 41 42 Frequently Asked Questions .................................................................................................................. 102 43 FAQ 4.1: How Will the Climate Change over the Next Twenty Years? .................................................... 102 44 FAQ 4.2: How Quickly Would We See the Effects of Reducing Carbon Dioxide Emissions? ................... 102 45 FAQ 4.3: At a given level of global warming, what are the spatial patterns of climate change? ................. 104 46 47 Acknowledgements................................................................................................................................ 106 48 References .............................................................................................................................................. 107 49 Figures.................................................................................................................................................... 148 50 51 Do Not Cite, Quote or Distribute 4-3 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Executive Summary 2 3 This chapter assesses simulations of future global climate change, spanning time horizons from the near term 4 (2021–2040), mid-term (2041–2060), and long term (2081–2100) out to the year 2300. Changes are assessed 5 relative to both the recent past (1995–2014) and the 1850–1900 approximation to the pre-industrial period. 6 7 The projections assessed here are mainly based on a new range of scenarios, the Shared Socio- 8 economic Pathways (SSPs) used in the Coupled Model Intercomparison Project Phase 6 (CMIP6). 9 Among the SSPs, the focus is on the five scenarios SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. 10 In the SSP labels, the first number refers to the assumed shared socio-economic pathway, and the second 11 refers to the approximate global effective radiative forcing (ERF) in 2100. Where appropriate, this chapter 12 also assesses new results from CMIP5, which used scenarios based on Representative Concentration 13 Pathways (RCPs). Additional lines of evidence enter the assessment, especially for change in globally 14 averaged surface air temperature (GSAT) and global mean sea level (GMSL), while assessment for changes 15 in other quantities is mainly based on CMIP6 results. Unless noted otherwise, the assessments assume that 16 there will be no major volcanic eruption in the 21st century. {1.6, 4.2.2, 4.3.2, 4.3.4, 4.6.2, BOX 4.1: Cross- 17 Chapter Box 4.1, Cross-Chapter Box 7.1, 9.6} 18 19 Temperature 20 21 Assessed future change in GSAT is, for the first time in an IPCC report, explicitly constructed by 22 combining scenario-based projections with observational constraints based on past simulated 23 warming, as well as an updated assessment of equilibrium climate sensitivity (ECS) and transient 24 climate response (TCR). Climate forecasts initialized using recent observations have also been used for the 25 period 2019–2028. The inclusion of additional lines of evidence has reduced the assessed uncertainty ranges 26 for each scenario. {4.3.1, 4.3.4, 4.4.1, 7.5} 27 28 In the near term (2021–2040), a 1.5°C increase in the 20-year average of GSAT, relative to the average 29 over the period 1850–1900, is very likely to occur in scenario SSP5-8.5, likely to occur in scenarios 30 SSP2-4.5 and SSP3-7.0, and more likely than not to occur in scenarios SSP1-1.9 and SSP1-2.6. The 31 threshold-crossing time is defined as the midpoint of the first 20-year period during which the average GSAT 32 exceeds the threshold. In all scenarios assessed here except SSP5-8.5, the central estimate of crossing the 33 1.5°C threshold lies in the early 2030s. This is about ten years earlier than the midpoint of the likely range 34 (2030–2052) assessed in the SR1.5, which assumed continuation of the then-current warming rate; this rate 35 has been confirmed in the AR6. Roughly half of the ten-year difference between assessed crossing times 36 arises from a larger historical warming diagnosed in AR6. The other half arises because for central estimates 37 of climate sensitivity, most scenarios show stronger warming over the near term than was assessed as 38 ‘current’ in SR1.5 (medium confidence). It is more likely than not that under SSP1-1.9, GSAT relative to 39 1850–1900 will remain below 1.6°C throughout the 21st century, implying a potential temporary overshoot 40 of 1.5°C global warming of no more than 0.1°C. If climate sensitivity lies near the lower end of the assessed 41 very likely range, crossing the 1.5°C warming threshold is avoided in scenarios SSP1-1.9 and SSP1-2.6 42 (medium confidence). {2.3.1, Cross-chapter Box 2.3, 3.3.1, 4.3.4, BOX 4.1:, 7.5} 43 44 By 2030, GSAT in any individual year could exceed 1.5°C relative to 1850–1900 with a likelihood 45 between 40% and 60%, across the scenarios considered here (medium confidence). Uncertainty in near- 46 term projections of annual GSAT arises in roughly equal measure from natural internal variability and model 47 uncertainty (high confidence). By contrast, near-term annual GSAT levels depend less on the scenario 48 chosen, consistent with the AR5 assessment. Forecasts initialized from recent observations simulate annual 49 GSAT changes for the period 2019–2028 relative to the recent past that are consistent with the assessed very 50 likely range (high confidence). {4.4.1, BOX 4.1:} 51 52 Compared to the recent past (1995–2014), GSAT averaged over the period 2081–2100 is very likely to 53 be higher by 0.2°C–1.0°C in the low-emission scenario SSP1-1.9 and by 2.4°C–4.8°C in the high- 54 emission scenario SSP5-8.5. For the scenarios SSP1-2.6, SSP2-4.5, and SSP3-7.0, the corresponding very 55 likely ranges are 0.5°C–1.5°C, 1.2°C–2.6°C, and 2.0°C–3.7°C, respectively. The uncertainty ranges for the Do Not Cite, Quote or Distribute 4-4 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 period 2081–2100 continue to be dominated by the uncertainty in ECS and TCR (very high confidence). 2 Emissions-driven simulations for SSP5-8.5 show that carbon-cycle uncertainty is too small to change the 3 assessment of GSAT projections (high confidence). {4.3.1, 4.3.4, 4.6.2, 7.5} 4 5 The CMIP6 models project a wider range of GSAT change than the assessed range (high confidence); 6 furthermore, the CMIP6 GSAT increase tends to be larger than in CMIP5 (very high confidence). 7 About half of the increase in simulated warming has occurred because higher climate sensitivity is more 8 prevalent in CMIP6 than in CMIP5; the other half arises from higher ERF in nominally comparable 9 scenarios (e.g., RCP8.5 and SSP5-8.5; medium confidence). In SSP1-2.6 and SSP2-4.5, ERF changes also 10 explain about half of the changes in the range of warming (medium confidence). For SSP5-8.5, higher 11 climate sensitivity is the primary reason behind the upper end of the warming being higher than in CMIP5 12 (medium confidence). {4.3.1, 4.3.4, 4.6.2, 7.5.6} 13 14 While high-warming storylines – those associated with GSAT levels above the upper bound of the 15 assessed very likely range – are by definition extremely unlikely, they cannot be ruled out. For SSP1-2.6, 16 such a high-warming storyline implies long-term (2081–2100) warming well above, rather than well 17 below, 2°C (high confidence). Irrespective of scenario, high-warming storylines imply changes in many 18 aspects of the climate system that exceed the patterns associated with the central estimate of GSAT changes 19 by up to more than 50% (high confidence). {4.3.4, 4.8} 20 21 It is virtually certain that the average surface warming will continue to be higher over land than over 22 the ocean and that the surface warming in the Arctic will continue to be more pronounced than the 23 global average over the 21st century. The warming pattern likely varies across seasons, with northern high 24 latitudes warming more during boreal winter than summer (medium confidence). Regions with increasing or 25 decreasing year-to-year variability of seasonal mean temperatures will likely increase in their spatial extent. 26 {4.3.1, 4.5.1, 7.4.4} 27 28 It is very likely that long-term lower-tropospheric warming will be larger in the Arctic than in the 29 global mean. It is very likely that global mean stratospheric cooling will be larger by the end of the 21st 30 century in a pathway with higher atmospheric CO2 concentrations. It is likely that tropical upper tropospheric 31 warming will be larger than at the tropical surface, but with an uncertain magnitude owing to the effects of 32 natural internal variability and uncertainty in the response of the climate system to anthropogenic forcing. 33 {4.5.1, 3.3.1.2} 34 35 Precipitation 36 37 Annual global land precipitation will increase over the 21st century as GSAT increases (high 38 confidence). The likely range of change in globally averaged annual land precipitation during 2081– 39 2100 relative to 1995–2014 is –0.2–4.7% in the low-emission scenario SSP1-1.9 and 0.9–12.9% in the 40 high-emission scenario SSP5-8.5, based on all available CMIP6 models. The corresponding likely ranges 41 are 0.0–6.6% in SSP1-2.6, 1.5–8.3% in SSP2-4.5, and 0.5–9.6% in SSP3-7.0. {4.3.1, 4.5.1, 4.6.1, 8.4.1} 42 43 Precipitation change will exhibit substantial regional differences and seasonal contrast as GSAT 44 increases over the 21st century (high confidence). As warming increases, a larger land area will experience 45 statistically significant increases or decreases in precipitation (medium confidence). Precipitation will very 46 likely increase over high latitudes and the tropical oceans, and likely increase in large parts of the monsoon 47 region, but likely decrease over large parts of the subtropics in response to greenhouse gas-induced warming. 48 Interannual variability of precipitation over many land regions will increase with global warming (medium 49 confidence). {4.5.1, 4.6.1, 8.4.1} 50 51 Near-term projected changes in precipitation are uncertain, mainly because of natural internal 52 variability, model uncertainty, and uncertainty in natural and anthropogenic aerosol forcing (medium 53 confidence). In the near term, no discernible differences in precipitation changes are projected between 54 different SSPs (high confidence). The anthropogenic aerosol forcing decreases in most scenarios, 55 contributing to increases in GSAT (medium confidence) and global-mean land precipitation (low Do Not Cite, Quote or Distribute 4-5 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 confidence). {4.3.1, 4.4.1, 4.4.4, 8.5} 2 3 In response to greenhouse gas-induced warming, it is likely that global land monsoon precipitation will 4 increase, particularly in the Northern Hemisphere, although Northern Hemisphere monsoon 5 circulation will likely weaken. In the long term (2081–2100), monsoon rainfall change will feature a north– 6 south asymmetry characterized by a greater increase in the Northern Hemisphere than in the Southern 7 Hemisphere and an east–west asymmetry characterized by an increase in Asian-African monsoon regions 8 and a decrease in the North American monsoon region (medium confidence). Near-term changes in global 9 monsoon precipitation and circulation are uncertain due to model uncertainty and internal variability such as 10 Atlantic Multi-decadal Variability and Pacific Decadal Variability (medium confidence). {4.4.1, 4.5.1, 8.4.1, 11 10.6.3} 12 13 It is likely that at least one large volcanic eruption will occur during the 21st century. Such an eruption 14 would reduce GSAT for several years, decrease global-mean land precipitation, alter monsoon 15 circulation, modify extreme precipitation, and change the profile of many regional climatic impact- 16 drivers. A low-likelihood, high-impact outcome would be several large eruptions that would greatly alter the 17 21st century climate trajectory compared to SSP-based Earth system model projections. {Cross-Chapter Box 18 4.1} 19 20 Large-scale Circulation and Modes of Variability 21 22 In the near term, the forced change in Southern Annular Mode in austral summer is likely to be 23 weaker than observed during the late 20th century under all five SSPs assessed. This is because of the 24 opposing influence in the near- to mid-term from stratospheric ozone recovery and increases in other 25 greenhouse gases on the Southern Hemisphere summertime mid-latitude circulation (high confidence). In the 26 near term, forced changes in the Southern Annular Mode in austral summer are therefore likely to be smaller 27 than changes due to natural internal variability. {4.3.3, 4.4.3} 28 29 In the long term, the Southern Hemisphere mid-latitude jet is likely to shift poleward and strengthen 30 under SSP5-8.5 relative to 1995–2014. This is likely to be accompanied by an increase in the Southern 31 Annular Mode index in all seasons relative to 1995–2014. For SSP1-2.6, CMIP6 models project no robust 32 change in the Southern Annular Mode index in the long term. It is likely that wind speeds associated with 33 extratropical cyclones will strengthen in the Southern Hemisphere storm track for SSP5-8.5. {4.5.1, 4.5.3} 34 35 The CMIP6 multi-model ensemble projects a long-term increase in the boreal wintertime Northern 36 Annular Mode index under the high-emission scenarios of SSP3-7.0 and SSP5-8.5, but regional 37 changes may deviate from a simple shift in the mid-latitude circulation. Substantial uncertainty and thus 38 low confidence remain in projecting regional changes in Northern Hemisphere jet streams and storm tracks, 39 especially for the North Atlantic basin in winter; this is due to large natural internal variability, the competing 40 effects of projected upper- and lower-tropospheric temperature gradient changes, and new evidence of 41 weaknesses in simulating past variations in North Atlantic atmospheric circulation on seasonal-to-decadal 42 timescales. One exception is the expected decrease in frequency of atmospheric blocking events over 43 Greenland and the North Pacific in boreal winter in SSP3-7.0 and SSP5-8.5 scenarios (medium confidence). 44 {4.5.1} 45 46 Near-term predictions and projections of the sub-polar branch of the Atlantic Multi-decadal 47 Variability (AMV) on the decadal timescale have improved in CMP6 models compared to CMIP5 48 (high confidence). This is likely to be related to a more accurate response to natural forcing in CMIP6 49 models. Initialization contributes to the reduction of uncertainty and to predicting subpolar sea surface 50 temperature. AMV influences on the nearby regions can be predicted over lead times of 5–8 years (medium 51 confidence). {4.4.3} 52 53 It is virtually certain that the El Niño–Southern Oscillation (ENSO) will remain the dominant mode of 54 interannual variability in a warmer world. There is no model consensus for a systematic change in 55 intensity of ENSO sea surface temperature (SST) variability over the 21st century in any of the SSP Do Not Cite, Quote or Distribute 4-6 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 scenarios assessed (medium confidence). However, it is very likely that ENSO rainfall variability, used for 2 defining extreme El Niños and La Niñas, will increase significantly, regardless of amplitude changes in 3 ENSO SST variability, by the second half of the 21st century in scenarios SSP2-4.5, SSP3-7.0, and SSP5- 4 8.5. {4.3.3, 4.5.3, 8.4.2} 5 6 Cryosphere and Ocean 7 8 Under the SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios, it is likely that the Arctic Ocean in September, 9 the month of annual minimum sea ice area, will become practically ice-free (sea ice area less than 1 10 million km2) averaged over 2081–2100 and all available simulations. Arctic sea ice area in March, the 11 month of annual maximum sea ice area, also decreases in the future under each of the considered scenarios, 12 but to a much lesser degree (in percentage terms) than in September (high confidence). {4.3.2} 13 14 Under the five scenarios assessed, it is virtually certain that global mean sea level (GMSL) will continue 15 to rise through the 21st century. For the period 2081–2100 relative to 1995–2014, GMSL is likely to rise 16 by 0.46–0.74 m under SSP3-7.0 and by 0.30–0.54 m under SSP1-2.6 (medium confidence). For the 17 assessment of change in GMSL, the contribution from land-ice melt has been added offline to the CMIP6- 18 simulated contributions from thermal expansion. {4.3.2. 9.6} 19 20 It is very likely that the cumulative uptake of carbon by the ocean and by land will increase through to 21 the end of the 21st century. Carbon uptake by land shows greater increases but with greater uncertainties 22 than for ocean carbon uptake. The fraction of emissions absorbed by land and ocean sinks will be smaller 23 under high emission scenarios than under low emission scenarios (high confidence). Ocean surface pH will 24 decrease steadily through the 21st century, except for SSP1-1.9 and SSP1-2.6 where values decrease until 25 around 2070 and then increase slightly to 2100 (high confidence). {4.3.2, 5.4} 26 27 Climate Response to Emission Reduction, Carbon Dioxide Removal, and Solar Radiation Modification 28 29 If strong mitigation is applied from 2020 onward as reflected in SSP1-1.9, its effect on 20-year trends 30 in GSAT would likely emerge during the near term (2021–2040), measured against an assumed non- 31 mitigation scenario such as SSP3-7.0 and SSP5-8.5. However, the response of many other climate 32 quantities to mitigation would be largely masked by internal variability during the near term, 33 especially on the regional scale (high confidence). The mitigation benefits for these quantities would 34 emerge only later during the 21st century (high confidence). During the near term, a small fraction of the 35 surface can show cooling under all scenarios assessed here, so near-term cooling at any given location is 36 fully consistent with GSAT increase (high confidence). Events of reduced and increased GSAT trends at 37 decadal timescales will continue to occur in the 21st century but will not affect the centennial warming (very 38 high confidence). {4.6.3, Cross-Chapter Box 3.1} 39 40 Because of the near-linear relationship between cumulative carbon emissions and GSAT change, the 41 cooling or avoided warming from carbon dioxide removal (CDR) is proportional to the cumulative 42 amount of CO2 removed by CDR (high confidence). The climate system response to net negative CO2 43 emissions is expected to be delayed by years to centuries. Net negative CO2 emissions due to CDR will not 44 reverse some climate change, such as sea level rise, at least for several centuries (high confidence). The 45 climate effect of a sudden and sustained CDR termination would depend on the amount of CDR-induced 46 cooling prior to termination and the rate of background CO2 emissions at the time of termination (high 47 confidence). {4.6.3, 5.5, 5.6} 48 49 Solar radiation modification (SRM) could offset some of the effects of anthropogenic warming on 50 global and regional climate, but there would be substantial residual and overcompensating climate 51 change at the regional scale and seasonal timescale (high confidence), and there is low confidence in 52 our understanding of the climate response to SRM, specifically at the regional scale. Since the AR5, 53 understanding of the global and regional climate response to SRM has improved, due to modelling work 54 with more sophisticated treatment of aerosol-based SRM options and stratospheric processes. Improved 55 modelling suggests that multiple climate goals could be met simultaneously. A sudden and sustained Do Not Cite, Quote or Distribute 4-7 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 termination of SRM in a high-emission scenario such as SSP5-8.5 would cause a rapid climate change (high 2 confidence). However, a gradual phase-out of SRM combined with emissions reductions and CDR would 3 more likely than not avoid larger rates of warming. {4.6.3} 4 5 Climate Change Commitment and Change Beyond 2100 6 7 Earth system modelling experiments since AR5 confirm that the zero CO2 emissions commitment (the 8 additional rise in GSAT after all CO2 emissions cease) is small (likely less than 0.3°C in magnitude) on 9 decadal time scales, but that it may be positive or negative. There is low confidence in the sign of the zero 10 CO2 emissions commitment. Consistent with SR1.5, the central estimate is taken as zero for assessments of 11 remaining carbon budgets for global warming levels of 1.5°C or 2°C. {4.7.2, 5.5.2}. 12 13 Overshooting specific global warming levels such as 2°C has effects on the climate system that persist 14 beyond 2100 (medium confidence). Under one scenario including a peak and decline in atmospheric CO2 15 concentration (SSP5-3.4-OS), some climate metrics such as GSAT begin to decline but do not fully reverse 16 by 2100 to levels prior to the CO2 peak (medium confidence). GMSL continues to rise in all models up to 17 2100 despite a reduction in CO2 to 2040 levels. {4.6.3, 4.7.1, 4.7.2} 18 19 Using extended scenarios beyond 2100, projections show likely warming by 2300, relative to 20 1850−1900, of 1.0°C−2.2°C for SSP1-2.6 and 6.6°C−14.1°C for SSP5-8.5. By 2300, warming under the 21 SSP5-3.4-OS overshoot scenario decreases from a peak around year 2060 to a level very similar to SSP1-2.6. 22 Precipitation over land continues to increase strongly under SSP5-8.5. GSAT projected for the end of the 23 23rd century under SSP2-4.5 (2.3−4.6°C) has not been experienced since the mid-Pliocene, about 3 million 24 years ago. GSAT projected for the end of the 23rd century under SSP5-8.5 (6.6−14.1°C) overlaps with the 25 range estimated for the Miocene Climatic Optimum (5°C−10°C) and Early Eocene Climatic Optimum 26 (10°C−18°C), about 15 and 50 million years ago, respectively (medium confidence). {2.3.1.1, 4.7.1} 27 Do Not Cite, Quote or Distribute 4-8 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 3 4.1 Scope and Overview of this Chapter 4 5 This chapter assesses simulations of future climate change, covering both near-term and long-term global 6 changes. The chapter assesses simulations of physical indicators of global climate change, such as global 7 surface air temperature (GSAT), global land precipitation, Arctic sea-ice area (SIA), and global mean sea 8 level (GMSL). Furthermore, the chapter covers indices and patterns of properties and circulation not only for 9 mean fields but also for modes of variability that have global significance. The choice of quantities to be 10 assessed is summarized in Cross-Chapter Box 2.2 and comprises a subset of the quantities covered in 11 Chapters 2 and 3. This chapter provides consistent coverage from near-term to long-term global changes and 12 provides the global reference for the later chapters covering important processes and regional change. 13 14 Essential input to the simulations assessed here is provided by future scenarios of concentrations or 15 anthropogenic emissions of radiatively active substances; the scenarios represent possible sets of decisions 16 by humanity, without any assessment that one set of decisions is more probable to occur than any other set 17 (see Chapter 1, Section 1.6). As in previous assessment reports, these scenarios are used for projections of 18 future climate using global atmosphere-ocean general circulation models (AOGCMs) and Earth system 19 models (ESMs, see Chapter 1, Section 1.5.3); the latter include representation of various biogeochemical 20 cycles such as the carbon cycle, the sulphur cycle, or ozone (e.g., Flato, 2011; Flato et al., 2013). This 21 chapter thus provides a comprehensive assessment of the future global climate response to different future 22 anthropogenic perturbations to the climate system. 23 24 Every projection assessment is conditioned on a particular forcing scenario. If sufficient evidence is 25 available, a detailed probabilistic assessment of a physical climate outcome can be performed for each 26 scenario separately. By contrast, there is no agreed-upon approach to assigning probabilities to forcing 27 scenarios, to the point that it has been debated whether such an approach can even exist (e.g., Grübler and 28 Nakicenovic, 2001; Schneider, 2001, 2002). Although there were some recent attempts to ascribe subjective 29 probabilities to scenarios (e.g., Ho et al., 2019; Hausfather and Peters, 2020), and although ‘feasibility’ along 30 different dimensions is an important concept in scenario research (see AR6 WGIII Chapter 3), the scenarios 31 used for the model-based projections assessed in this chapter do not come with statements about their 32 likelihood of actually unfolding in the future. Therefore, it is usually not possible to combine responses to 33 individual scenarios into an overall probabilistic statement about expected future climate. Exceptions to this 34 limit in the assessment are possible only under special circumstances, such as for some statements about 35 near-term climate changes that are largely independent of the scenario chosen (e.g., Section 4.4.1). Beyond 36 this, no combination of responses to different scenarios can be assessed in this chapter but may be possible in 37 future assessments. 38 39 A central element of this chapter is a comprehensive assessment of the sources of uncertainty of future 40 projections (see Chapter 1, Section 1.4.3). Uncertainty can be broken down into scenario uncertainty, model 41 uncertainty involving model biases, uncertainty in simulated effective radiative forcing and model response, 42 and the uncertainty arising from internal variability (Cox and Stephenson, 2007; Hawkins and Sutton, 2009). 43 An additional source of projection uncertainty arises from possible future volcanic eruptions and future solar 44 variability. Assessment of uncertainty relies on multi-model ensembles such as the Coupled Model 45 Intercomparison Project Phase 6 (CMIP6, Eyring et al., 2016), single-model initial-condition large 46 ensembles (e.g., Kay et al., 2015; Deser et al., 2020), and ensembles initialized from the observed climate 47 state (decadal predictions, e.g., Smith et al., 2013a; Meehl et al., 2014a; Boer et al., 2016; Marotzke et al., 48 2016). Ensemble evaluation methods include assessment of model performance and independence (e.g., 49 Knutti et al., 2017; Boe, 2018; Abramowitz et al., 2019); emergent and other observational constraints (e.g., 50 Allen and Ingram, 2002; Hall and Qu, 2006; Cox et al., 2018); and the uncertainty assessment of equilibrium 51 climate sensitivity and transient climate response in Chapter 7. Ensemble evaluation is assessed in Box 4.1 52 through the inclusion of lines of evidence in addition to the projection ensembles, including implications for 53 potential model weighting. 54 55 The uncertainty assessment in this chapter builds on one particularly noteworthy advance since the IPCC Do Not Cite, Quote or Distribute 4-9 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Fifth Assessment Report (AR5). Internal variability, which constitutes irreducible uncertainty over much of 2 the time horizon considered here (Hawkins et al., 2016; Marotzke, 2019), can be better estimated in models 3 even under a changing climate through the use of large initial-condition ensembles (Kay et al., 2015). For 4 many climate quantities and compared to the forced climate change signal, internal variability is dominant in 5 any individual realization – including the one that will unfold in reality – in the near term (Kirtman et al., 6 2013; Marotzke and Forster, 2015), is substantial in the mid-term, and is still recognizable in the long term in 7 many quantities (Deser et al., 2012a; Marotzke and Forster, 2015). This chapter will use the strengthend 8 information on internal variability throughout. 9 10 The expanded treatment of uncertainty allows this chapter a more comprehensive assessment of the benefits 11 from mitigation than in previous IPCC reports, as well as the climate response to Carbon Dioxide Removal 12 (CDR) and Solar Radiation Modification (SRM), and how to detect them against the backdrop of internal 13 variability. Important advances have been made in the detection and attribution of mitigation, CDR, and 14 SRM (Bürger and Cubasch, 2015; Lo et al., 2016; Ciavarella et al., 2017); exploring the ‘time of emergence’ 15 (ToE; see Annex VII: Glossary) of responses to assumed emissions reductions (Tebaldi and Friedlingstein, 16 2013)(Samset et al., 2020) and the attribution of decadal events to forcing changes that reflect emissions 17 reductions (Marotzke, 2019; Spring et al., 2020)(McKenna et al., 2021). 18 19 The question of the potential crossing of thresholds relative to global temperature goals (Geden and 20 Loeschel, 2017) is intimately related to the benefits of mitigation; a prerequisite is an assessment of how 21 robustly magnitudes of warming can be defined (Millar et al., 2017). This chapter provides an update to the 22 IPCC Special Report on Global Warming of 1.5°C (SR1.5, Masson-Delmotte et al., 2018) and constitutes a 23 reference point for later chapters and AR6 WGIII on the effects of mitigation, including a robust uncertainty 24 assessment. 25 26 27 [START FIGURE 4.1 HERE] 28 29 Figure 4.1: Visual abstract of Chapter 4. The chapter outline and a quick guide for key topics and corresponding 30 subsections are provided. 31 32 [END FIGURE 4.1 HERE] 33 34 35 The chapter is organized as follows (see Figure 4.1). After Section 4.2 on the methodologies used in the 36 assessment, Section 4.3 assesses projected changes in key global climate indicators throughout the 21st 37 century, relative to the period 1995–2014, which comprises the last twenty years of the historical simulations 38 of CMIP6 (Eyring et al., 2016) and hence the most recent past simulated with the observed atmospheric 39 composition. The global climate indicators assessed include GSAT, global land precipitation, Arctic sea-ice 40 area (SIA), global mean sea level (GMSL), the Atlantic Meridional Overturning Circulation (AMOC), global 41 mean ocean surface pH, carbon uptake by land and ocean, the global monsoon, the Northern and Southern 42 Annular Modes (NAM and SAM), and the El Niño–Southern Oscillation (ENSO). Differently from the 43 assessment for changes in other quantities only based on the range of CMIP6 projections, additional lines of 44 evidence enter the assessment for GSAT and GMSL change. For most results and figures based on CMIP6, 45 one realization from each model (the first of the uploaded set) is used. Section 4.3 finally synthesizes the 46 assessment of GSAT change using multiple lines of evidence in addition to the CMIP6 projection 47 simulations. 48 49 Section 4.4 covers near-term climate change, defined here as the period 2021–2040 and taken relative to the 50 period 1995–2014. Section 4.4 focuses on global and large-scale climate indicators, including precipitation 51 and circulation indices and selected modes of variability (see Cross-chapter Box 2.2 and Annex AIV), as 52 well as on the spatial distribution of warming. The potential roles of short-lived climate forcers (SLCFs) and 53 volcanic eruptions on near-term climate change are also discussed. Section 4.4 synthesizes information from 54 initialized predictions and non-initialized projections for the near-term change. 55 Do Not Cite, Quote or Distribute 4-10 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Section 4.5 then covers mid-term and long-term climate change, defined here as the periods 2041–2060 and 2 2081–2100, respectively, again relative to the period 1995–2014. The mid-term period is thus chosen as the 3 twenty-year period following the short-term period and straddling the mid-century point, year 2050; it is 4 during the mid-term that differences between different scenarios are expected to emerge against internal 5 variability. The long-term period is defined, as in the AR5, as the twenty-year period at the end of the 6 century. Section 4.5 assesses the same set of indicators as Section 4.4, as well as changes in internal 7 variability and in large-scale patterns, both of which are expected to emerge in the mid- to long-term. The 8 chapter sub-division according to time slices (near term, mid-term, and long term) is thus to a large extent 9 motivated by the different roles that internal variability plays in each period, compared to the expected 10 forced climate-change signal. 11 12 Section 4.6 assesses the climate implications of climate policies, as simulated with climate models. First, 13 Section 4.6 assesses patterns of climate change expected for various levels of GSAT rise including 1.5°C, 14 2°C, 3°C, and 4°C, compared to the approximation to the pre-industrial period 1850–1900 to facilitate 15 immediate connection to the SR1.5 and the temperature goals specified in the Paris Agreement (UNFCCC, 16 2016). Section 4.6 continues with climate goals, overshoot, and path-dependence, as well as the climate 17 response to mitigation, CDR, and SRM. Section 4.6 also covers the consistency between RCPs and SSPs. 18 19 Section 4.7 assesses very long-term changes in selected global climate indicators, from 2100 to 2300. 20 Section 4.7 continues with climate-change commitment and the potential for irreversibility and abrupt 21 climate change. The chapter concludes with Section 4.8 on the potential for low-probability–high-impact 22 changes, followed by answers to three frequently asked questions (FAQs). 23 24 25 4.2 Methodology 26 27 4.2.1 Models, Model Intercomparison Projects, and Ensemble Methodologies 28 29 Similar to the approach used in AR5 (Flato et al., 2013), the primary lines of evidence of this chapter are 30 comprehensive climate models (atmosphere-ocean general circulation models, AOGCMs) and Earth system 31 models (ESMs); ESMs differ from AOGCMs by including representations of various biogeochemical cycles. 32 We also build on results from Earth system models of intermediate complexity (EMICs, Claussen et al., 33 2002; Eby et al., 2013) and other types of models where appropriate. This chapter focuses on a particular set 34 of coordinated multi-model experiments known as model intercomparison projects (MIPs). These 35 frameworks recommend and document standards for experimental design for running AOGCMs and ESMs 36 to minimise the chance of differences in results being misinterpreted. CMIP is an activity of the World 37 Climate Research Programme (WCRP), and the latest phase is CMIP6 (Eyring et al., 2016). To establish 38 robustness of results, it is vital to assess the performance of these models in terms of mean state, variability, 39 and the response to external forcings. That evaluation has been undertaken using the CMIP6 ‘Diagnostic, 40 Evaluation and Characterization of Klima’ (DECK) and historical simulations in AR6 Chapter 3, which 41 concludes that there is high confidence that the CMIP6 multi-model mean captures most aspects of observed 42 climate change well (Chapter 3, Section 3.8.3.1). 43 44 This chapter draws mainly on future projections referenced both against the period 1850–1900 and the recent 45 past, 1995–2014, performed primarily under ScenarioMIP (O’Neill et al., 2016). This allows us to assess 46 both dimensions of integration across scenarios (Section 4.3) and global warming levels (Section 4.6) as 47 discussed in Chapter 1, Section 1.6. Other MIPs also target future scenarios with a focus on specific 48 processes or feedbacks and are summarised in Table 4.1. 49 50 51 [START TABLE 4.1 HERE] 52 53 Table 4.1: Model Intercomparison Projects (MIPs) utilized in Chapter 4. 54 MIP / experiment Usage Chapter/Section Reference Do Not Cite, Quote or Distribute 4-11 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI DECK, 1%, 4 • CO2 Diagnosing climate Assessed in Ch7 (Eyring et al., 2016) sensitivity ECS and TCR used in GSAT assessment CMIP6 Historical Evaluation, baseline Assessed in Ch3 (Eyring et al., 2016) Used in chapter 4 to cover reference period ScenarioMIP Future projections throughout Ch.4 (O’Neill et al., 2016) AerChemMIP Aerosols and trace gases 4.4.4 (Collins et al., 2017) C4MIP CO2 emissions driven 4.3.1 (Jones et al., 2016b) simulations CDRMIP Carbon Dioxide Removal 4.6.3 (Keller et al., 2018) GeoMIP Solar Radiation 4.6.3 (Kravitz et al., 2011) Modification PDRMIP Forcing dependence of 4.5.1 (Myhre et al., 2017) precipitation SIMIP Sea ice assessment 4.3 (Notz et al., 2016) ZECMIP Zero emissions 4.7.1 (Jones et al., 2019a) commitment CMIP5 RCP scenario assessment 4.6.2, 4.7.1 (Taylor et al., 2012) 1 2 [END TABLE 4.1 HERE] 3 4 5 Multi-model ensembles provide the central focus of projection assessment. While single-model experiments 6 have great value for exploring new results and theories, multi-model ensembles additionally underpin the 7 assessment of the robustness, reproducibility, and uncertainty attributable to model internal structure and 8 processes variability (Hawkins and Sutton, 2009) (see Section 4.2.5). Techniques underlying the 9 combination of evaluation and weighting that are applied in this chapter are synthesized in Box 4.1. 10 11 Climate model simulations can be performed in either ‘concentration-driven’ or ‘emissions-driven’ 12 configuration reflecting whether the CO2 concentration is prescribed to follow a pre-defined pathway or is 13 simulated by the Earth system models in response to prescribed emissions of CO2 (see Box 6.4 in Ciais et al., 14 2013). The majority of CMIP6 experiments are conducted in concentration-driven configurations in order to 15 enable models without a fully interactive carbon cycle to perform them, and throughout most of this chapter 16 we present results from those simulations unless otherwise stated. Concentrations of other greenhouse gases 17 are always prescribed. However, the SSP5-8.5 scenario has also been performed in emissions-driven 18 configuration (‘esm-ssp585’) by ten ESMs, and in Section 4.3.1.1 we assess the impact on simulated climate 19 over the 21st century. 20 21 Internal variability complicates the identification of forced climate signals, especially when considering 22 regional climate signals over short timescales (up to a few decades), such as local trends over the satellite era 23 (Hawkins and Sutton, 2009; Deser et al., 2012a; Xie et al., 2015; Lovenduski et al., 2016; Suárez-Gutiérrez 24 et al., 2017). Large initial-condition ensembles, where the same model is run repeatedly under identical 25 forcing but with initial conditions varied through small perturbations or by sampling different times of a pre- 26 industrial control run, have substantially grown in their use since the AR5 (Deser et al., 2012a; Kay et al., 27 2015; Rodgers et al., 2015; Hedemann et al., 2017; Stolpe et al., 2018; Maher et al., 2019a). Such large 28 ensembles have shown potential to quantify uncertainty due to internal variability (Hawkins et al., 2016; 29 McCusker et al., 2016; Sigmond and Fyfe, 2016; Lehner et al., 2017; McKinnon et al., 2017; Marotzke, 30 2019) and thereby extract the forced signal from the internal variability, which can be calibrated against 31 observational data to improve the reliability of probabilistic climate projections over the near and mid-term 32 (O’Reilly et al., 2020). Moreover, they allow the investigation of forced changes in internal variability (e.g., 33 Maher et al., 2018). A key assumption is that a given model skilfully represents internal variability; 34 structural uncertainty is not accounted for. 35 36 A complementary approach that represents structural uncertainty in a given model is stochastic physics 37 (Berner et al., 2017). The approach has proven useful in representing structural uncertainty on seasonal Do Not Cite, Quote or Distribute 4-12 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 climate timescales (Weisheimer et al., 2014; Batté and Doblas-Reyes, 2015; MacLachlan et al., 2015). 2 Stochastic physics can markedly improve the internal variability in a given model (Dawson and Palmer, 3 2015; Wang et al., 2016; Christensen et al., 2017; Davini et al., 2017; Watson et al., 2017; Strømmen et al., 4 2018; Yang et al., 2019). Stochastic physics can also correct long-standing mean-state biases (Sanchez- 5 Gomez et al., 2016) and can influence the predicted climate sensitivity (Christensen and Berner, 2019; 6 Strommen et al., 2019; Meccia et al., 2020). 7 8 Perturbed-physics ensembles (Murphy et al., 2004) are also used to systematically account for parameter 9 uncertainty in a given model. Uncertain model parameters are identified and ranges in their values selected 10 that conform to emergent observational constraints (see Chapter 1, Section 1.5.4.2). These parameters are 11 then changed between ensemble members to sample the effect of parameter uncertainty on climate (Piani et 12 al., 2005; Sexton et al., 2012; Johnson et al., 2018; Regayre et al., 2018). It is possible to weight ensemble 13 members according to some performance metric or emergent constraint (e.g., Fasullo et al., 2015; Section 14 1.5.4.7) to improve the ensemble distribution (see also Box 4.1). 15 16 17 4.2.2 Scenarios 18 19 The AR5 drew heavily on four main scenarios, known as Representative Concentration Pathways (RCPs: 20 Meinshausen et al., 2011; van Vuuren et al., 2011), and simulation results from CMIP5 (Taylor et al., 2012, 21 see Section 4.2.1). The RCPs were labelled by the approximate radiative forcing reached at the year 2100, 22 going from 2.6, 4.5, 6.0 to 8.5 W m-2. 23 24 This chapter draws on model simulations from CMIP6 (Eyring et al., 2016) using a new range of scenarios 25 based on Shared Socio-economic Pathways (SSPs; O’Neill et al., 2016). The set of SSPs is described in 26 detail in Chapter 1 (see Section 1.6) and recognizes that global radiative forcing levels can be achieved by 27 different pathways of CO2, non-CO2 greenhouse gases (GHGs), aerosols (Amann et al., 2013; Rao et al., 28 2017), and land use; the set of SSPs therefore establishes a matrix of global forcing levels and socio- 29 economic storylines. ScenarioMIP (O’Neill et al., 2016) identifies four priority (tier-1) scenarios that 30 participating modelling groups are asked to perform, SSP1-2.6 for sustainable pathways, SSP2-4.5 for 31 middle-of-the-road, SSP3-7.0 for regional rivalry, and SSP5-8.5 for fossil-fuel-rich development. This 32 chapter focuses its assessment on these, plus the SSP1-1.9 scenario, which is directly relevant the assessment 33 of the 1.5°C Paris Agreement goal. Further, this chapter discusses these scenarios and their extensions past 34 2100 in the context of the very long-term climate change in Section 4.7.1. Projections of short-lived climate 35 forcers (SLCFs) are assessed in more detail in Chapter 6 (Section 6.7). 36 37 In presenting results and evidence, this chapter tries to be as comprehensive as possible. In tables we show 38 multi-model mean change and 5⎼95% range for all five SSPs, while in time series figures we show multi- 39 model mean change for all five SSPs but for clarity 5⎼95% range only for SSP1-2.6 and SSP3-7.0. Where 40 maps are presented, due to space restrictions we focus on showing multi-model mean change for SSP1-2.6 41 and SSP3-7.0. SSP1-2.6 is preferred over SSP1-1.9 because the latter has far fewer simulations available. 42 The high-end scenarios RCP8.5 or SSP5-8.5 have recently been argued to be implausible to unfold (e.g., 43 (Hausfather and Peters, 2020); see Chapter 3 of the AR6 WGIII). However, where relevant we show results 44 for SSP5-8.5, for example to enable backwards compatibility with AR5, for comparison between emission- 45 driven and concentration-driven simulations, and because there is greater data availability of daily output for 46 SSP5-8.5. When presenting low-likelihood high-warming storylines we also show results from the high-end 47 SSP5-8.5 scenario. 48 49 ScenarioMIP simulations include advances in techniques to better harmonize with historical forcings relative 50 to CMIP5. For example, projected changes in the solar cycle include long-term modulation rather than a 51 repeating solar cycle (Matthes et al., 2017). Background natural aerosols are ramped down to an average 52 historical level used in the control simulation by 2025 and background volcanic forcing is ramped up from 53 the value at the end of the historical simulation period (2015) over 10 years to the same constant value 54 prescribed for the piControl simulations in the DECK, and then kept fixed – both changes are intended to 55 avoid inconsistent model treatment of unknowable natural forcing to affect the near-term projected warming. Do Not Cite, Quote or Distribute 4-13 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 Complete backward comparability between CMIP5 and CMIP6 scenarios cannot be established for detailed 3 regional assessments, because the SSP scenarios include regional forcings ⎼ especially from land use and 4 aerosols ⎼ that are different from the CMIP5 RCPs. Even at a global level, a quantitative comparison is 5 challenging between corresponding SSP and RCP radiative forcing levels due to differing contributions to 6 the forcing (Meinshausen et al., 2020) and evidence of differing model responses (Wyser et al., 2020) 7 (Section 4.6.2.2). The RCP scenarios assessed in the AR5 all showed similar, rapid reductions in SLCFs and 8 emissions of SLCF precursor species over the 21st century; the CMIP5 projections hence did not sample a 9 wide range of possible trajectories for future SLCFs (Chuwah et al., 2013). The SSP scenarios assessed in 10 the AR6 offer more scope to explore SLCF pathways as they sample a broader range of air quality policy 11 options (Gidden et al., 2019) and relationships of CO2 to non-CO2 greenhouse gases (Meinshausen et al., 12 2020). Section 4.6.2.2 assesses RCP and SSP differences. Other MIPs (see Section 4.2.1) have been designed 13 to explicitly explore some of the implications of the different socio-economic storylines for a given radiative 14 forcing level. 15 16 17 4.2.3 Sources of Near-Term Information 18 19 This subsection describes the three main sources of near-term information used in Chapter 4. These are 1) 20 the projections from the CMIP6 multi-model ensemble introduced in Section 4.2.1 (Eyring et al., 2016; 21 O’Neill et al., 2016), 2) observationally constrained projections (Gillett et al., 2013; Stott et al., 2013), and 3) 22 the initialized predictions contributed to CMIP6 from the Decadal Climate Prediction Project (DCPP, Boer et 23 al., 2016). The projections under 1) and the observational constraints under 2) are used for all time horizons 24 considered in this chapter, whereas the initialized predictions under 3) are relevant only in the near term. 25 26 Observationally constrained projections (Gillett et al., 2013, 2021; Shiogama et al., 2016; Ribes et al., 2021) 27 use detection and attribution methods to attempt to reach consistency between observations and models and 28 thus provide improved projections of near-term change. Notable advances have been made since the AR5, 29 for example the ability to observationally constrain estimates of Arctic sea-ice loss for global warming of 30 1.5°C, 2.0°C, and 3.0°C above pre-industrial levels (Screen and Williamson, 2017; Jahn, 2018; Screen, 31 2018; Sigmond et al., 2018). There is high confidence that these approaches can reduce the uncertainties 32 involved in such estimates. 33 34 The AR5 was the first IPCC report to assess decadal climate predictions initialized from the observed 35 climate state (Kirtman et al., 2013), and assessed with high confidence that these predictions exhibit positive 36 skill for near-term average surface temperature information, globally and over large regions, for up to ten 37 years. Substantially more experience in producing initialized decadal predictions has been gained since the 38 AR5; the remainder of this subsection assesses the advances made. 39 40 Because the ‘memory’ that potentially enables prediction of multi-year to decadal internal variability resides 41 mainly in the ocean, some systems initialize the ocean state only (e.g., Müller et al., 2012; Yeager et al., 42 2018), whereas others incorporate observed information in the initial atmospheric states (e.g., Pohlmann et 43 al., 2013; Knight et al., 2014) or other non-oceanic drivers that provide further sources of predictability 44 (Alessandri et al., 2014; Weiss et al., 2014; Bellucci et al., 2015a). 45 46 Ocean initialization techniques generally use one of two strategies. Under full-field initialization, estimates 47 of observed climate states are represented directly on the model grid. A potential drawback is that 48 predictions initialized using the full-field approach tend to drift toward the biased climate preferred by the 49 model (Smith et al., 2013a; Bellucci et al., 2015c; Sanchez-Gomez et al., 2016; Kröger et al., 2018; Nadiga 50 et al., 2019). Such drifts can be as large as, or larger than, the climate anomaly being predicted and may 51 therefore obscure predicted climate anomalies (Kröger et al., 2018) unless corrected for through post- 52 processing. By contrast, anomaly initialization reduces drifts by adding observed anomalies (i.e., deviations 53 from mean climate) to the mean model climate (Pohlmann et al., 2013; Smith et al., 2013a; Thoma et al., 54 2015b; Cassou et al., 2018), but has the disadvantage that the model state is then further from the real world Do Not Cite, Quote or Distribute 4-14 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 from the start of the prediction. For both approaches, unrealistic features in the observation data used for 2 initialization may induce unrealistic transient behavior (Pohlmann et al., 2017; Teng et al., 2017; Nadiga et 3 al., 2019), and non-linearity can reduce forecast skill (Chikamoto et al., 2019). As yet, neither of the 4 initialization strategies has been clearly shown to be superior (Hazeleger et al., 2013; Magnusson et al., 5 2013; Smith et al., 2013a; Marotzke et al., 2016), although such comparisons may be sensitive to the model, 6 region, and details of the initialization and forecast assessment procedures considered (Polkova et al., 2014; 7 Bellucci et al., 2015c). 8 9 There is also a wide range of techniques employed to assimilate observed information into models in order to 10 generate suitable initial conditions (Polkova et al., 2019). These range in complexity from simple relaxation 11 towards observed time series of sea surface temperature (SST) (Mignot et al., 2016) or wind stress anomalies 12 (Thoma et al., 2015a, 2015b), to relaxation toward three-dimensional ocean and sometimes atmospheric state 13 estimates from various sources (e.g., Pohlmann et al., 2013; Knight et al., 2014; Dunstone et al., 2016), or 14 hybrid relaxation combining surface and tri-dimensional restoring as function of ocean basins and depth 15 (Sanchez-Gomez et al., 2016), to sophisticated data assimilation methods such as the ensemble Kalan filter 16 (Nadiga et al., 2013; Counillon et al., 2014, 2016; Msadek et al., 2014; Karspeck et al., 2015; Brune et al., 17 2018; Cassou et al., 2018; Polkova et al., 2019) the four-dimensional ensemble-variational hybrid data 18 assimilation (He et al., 2017, 2020) and the initialization of sea ice (Guemas et al., 2016; Kimmritz et al., 19 2018). In addition, decadal predictions necessarily consist of ensembles of forecasts to quantify uncertainty 20 as discussed in Section 4.2.1. A common way to generate an ensemble is through sets of initial conditions 21 containing small variations that lead to different subsequent climate trajectories. A variety of methods and 22 assumptions has been employed to generate and filter initial-condition ensembles for decadal prediction 23 (e.g., Marini et al., 2016; Kadow et al., 2017). As yet, there is no clear consensus as to which initialization 24 and ensemble generation techniques are most effective, and evaluations of their comparative performance 25 within a single modelling framework are needed (Cassou et al., 2018). 26 27 A consequence of model imperfections and resulting model systematic errors is that estimates of these errors 28 must be removed from the prediction to isolate the predicted climate anomaly and the phase of the decadal 29 modes of climate variability (see Annex IV, Sections AIV.2.6 and AIV.2.7 and Section 4.4.3.5 and 4.4.3.6). 30 Because of the tendency for systematic drifts to occur following initialization, bias corrections generally 31 depend on time since the start of the forecast, often referred to as lead time. In practice, the lead-time- 32 dependent biases are calculated using ensemble retrospective predictions, also known as hindcasts, and 33 recommended basic procedures for such corrections are provided in previous studies (Goddard et al., 2013; 34 Boer et al., 2016). The biases are also dynamically corrected during hindcasts and predictions by 35 incorporating the multi-year monthly mean analysis increments from the initialization into the initial 36 condition at each integration step (Wang et al., 2013b). Besides mean climate as a function of lead time, 37 further aspects of decadal predictions may be biased, such as the modes of variability (see Annex IV and 38 Chapter 3, Section 3.7) upon which drift patterns are projecting (Sanchez-Gomez et al., 2016), and additional 39 correction procedures have thus been proposed to remove biases in representing long-term trends (Kharin et 40 al., 2012; Kruschke et al., 2016; Balaji et al., 2018; Pasternack et al., 2018), as well as more general 41 dependences of drift on initial conditions (Fučkar et al., 2014; Pasternack et al., 2018; Nadiga et al., 2019). 42 43 Many skill measures exist that describe different aspects of the correspondence between predicted and 44 observed conditions, and no single metric should be considered exclusively. Important aspects of forecast 45 performance captured by different skill measures include: 1) the ability to predict the sign and phases of the 46 main modes of decadal variability and their regional fingerprint through teleconnections; 2) the typical 47 magnitude of differences between predicted and observed values, forecast reliability and resolution (Corti et 48 al., 2012); and 3) whether the forecast ensemble appropriately represents uncertainty in the predictions. A 49 framework for skill assessment that encompasses each of these aspects of forecast quality has been proposed 50 (Goddard et al., 2013). A new, process-based method to assess forecast skill in decadal predictions is to 51 analyse how well a specific mechanism is represented at each lead time (Mohino et al., 2016). 52 53 One additional aspect of forecast quality assessment is that estimated skill can be degraded by errors in 54 observational datasets used for verification, in addition to errors in the predictions (Massonnet et al., 2016; 55 Ferro, 2017; Karspeck et al., 2017; Juricke et al., 2018). This suggests that skill may tend to be Do Not Cite, Quote or Distribute 4-15 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 underestimated, particularly for climate variables whose observational uncertainties are relatively large, and 2 that the predictions themselves may prove useful for assessing the quality of observational datasets 3 (Massonnet, 2019). 4 5 Skill assessments have shown that initialized predictions can out-perform their uninitialized counterparts 6 (Doblas-Reyes et al., 2013; Meehl et al., 2014; Bellucci et al., 2015b; Smith et al., 2018; Yeager et al., 2018; 7 Smith et al., 2019b), although such comparisons are sensitive to the region and variable considered, multi- 8 model predictions are generally more skilful than individual models (Doblas-Reyes et al., 2013; Smith et al., 9 2013b, 2019b). Considerable skill, especially for temperature, can be attributed to external forcings such as 10 changes in GHG, aerosol concentrations, and volcanic eruptions. On a global scale, this contribution to skill 11 has been found to exceed that from the prediction of internal variability except in the early stages (about one 12 year for global SST) of the forecast (Corti et al., 2015)(Sospedra-Alfonso and Boer, 2020)(Bilbao et al., 13 2021), though idealized potential skill measures and observations-based studies suggest that improving the 14 prediction of internal variability could extend this crossover to longer lead times (Boer et al., 2013; Årthun et 15 al., 2017). In some cases, part of the skill arises from the ability of initialized predictions to capture observed 16 transitions of major modes of climate variability (Meehl et al., 2016) such as the Pacific Decadal Variability 17 (PDV) and the Atlantic Multidecadal Variability (AMV) (see Sections 4.4.3.5 and 4.4.3.6, and Annex IV, 18 Sections AIV.2.6 and AIV.2.7). 19 20 Initialized predictions of near-surface temperature are particularly skilful over the North Atlantic, a region of 21 high potential and realised predictability (Keenlyside et al., 2008; Pohlmann et al., 2009; Boer et al., 2013; 22 Yeager and Robson, 2017). Much of this predictability is associated with the North Atlantic subpolar gyre 23 (Wouters et al., 2013), where skill in predicting ocean conditions is typically high (Hazeleger et al., 2013; 24 Brune and Baehr, 2020) and shifts in ocean temperature and salinity potentially affecting surface climate can 25 be predicted up to several years in advance (Robson et al., 2012; Hermanson et al., 2014), although such 26 assessments remain challenging due to incomplete knowledge of the state of the ocean during the hindcast 27 evaluation periods (Menary and Hermanson, 2018). A substantial improvement of the sub-polar gyre SST 28 prediction is found in CMIP6 models, which is attributed to a more accurate response to the AMOC-related 29 delayed response to volcanic eruptions (Borchert et al., 2021) (see Section 4.4.3). A significant 30 improvement GSAT prediction skill is also found over some land regions including East Asia (Monerie et 31 al., 2018), Eurasia (Wu et al., 2019), Europe (Müller et al., 2012; Smith et al., 2019b) and the Middle East 32 (Smith et al., 2019b). 33 34 Skill for multi-year to decadal precipitation forecasts is generally much lower than for temperature, although 35 one exception is Sahel rainfall (Sheen et al., 2017), due to its dependence on predictable variations in North 36 Atlantic SST through teleconnections (Martin and Thorncroft, 2014a) (Annex IV). Predictive skill on 37 decadal timescales is found for extratropical storm-tracks and storm density (Kruschke et al., 2016; Schuster 38 et al., 2019a), atmospheric blocking (Schuster et al., 2019b; Athanasiadis et al., 2020), the Quasi-Biennial 39 Oscillation (QBO) (Scaife et al., 2014; Pohlmann et al., 2019) and over the tropical oceans (tropical trans- 40 basin variability) (Chikamoto et al., 2015). In addition, decadal predictions with large ensemble sizes are 41 able to predict multi-annual temperature (Peters et al., 2011)(Sienz et al., 2016)(Borchert et al., 2019) 42 (Sospedra-Alfonso and Boer, 2020), precipitation (Yeager et al., 2018; Smith et al., 2019b), and atmospheric 43 circulation (Smith et al., 2020) anomalies over certain land regions, although the ensemble-mean magnitudes 44 are much weaker than observed. This discrepancy may be symptomatic of an apparent deficiency in climate 45 models that causes some predictable signal, such as that associated to the North Atlantic Oscillation (NAO) 46 (see Section AIV.2.1), to be much weaker than in nature (Eade et al., 2014; Scaife and Smith, 2018; 47 Strommen and Palmer, 2019; Smith et al., 2020), while others, such as that linked to the Southern Annular 48 Mode (SAM) (see Section AIV.2.2), are more consistent with observations (Byrne et al., 2019) 49 50 Evidence is accumulating that additional properties of the Earth system relating to ocean variability may be 51 skilfully predicted on multi-annual timescales. These include levels of Atlantic hurricane activity (Smith et 52 al., 2010; Caron et al., 2017), winter sea-ice in the Arctic (Onarheim et al., 2015; Dai et al., 2020), drought 53 and wildfire (Chikamoto et al., 2017; Paxian et al., 2019; Solaraju-Murali et al., 2019), and variations in the 54 ocean carbon cycle including CO2 uptake (Li et al., 2016b, 2019; Lovenduski et al., 2019; Fransner et al., 55 2020) and chlorophyll (Park et al., 2019). Do Not Cite, Quote or Distribute 4-16 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 In summary, despite challenges (Cassou et al., 2018), there is high confidence that initialized predictions 3 contribute information to near-term climate change for some regions over multiannual to decadal timescales. 4 Furthermore, there are indications that initialized predictions can constrain near-term projections (Befort et 5 al., 2020). The clearest improvements through initialization are seen in the North Atlantic and related 6 phenomena such as hurricane frequency, Sahel and European rainfall. By contrast, there is medium or low 7 confidence that uncertainty is reduced for other climate variables. 8 9 10 4.2.4 Pattern Scaling 11 12 Projected climate change is typically represented in this chapter for specific future periods. One important 13 source of uncertainty in projections presented for fixed future epochs (time-slabs/time-slices) is the 14 underlying scenario used; another is the structural uncertainty associated with model climate sensitivity. 15 Presenting projections and associated measures of uncertainty for specific periods (see Sections 4.4 and 4.5) 16 remains the most widely applied methodology towards informing climate change impact studies. It is 17 becoming increasingly important from the perspective of climate change and mitigation policy, however, to 18 present projections also as a function of the change in global mean temperature (i.e., global warming levels, 19 GWLs). They are expressed either in terms of changes of global mean surface temperature (GMST) or 20 GSAT (see Chapter 1, Section 1.6.2 and Cross-Chapter Box 2.3). For example, the IPCC SR1.5 (Hoegh- 21 Guldberg et al., 2018) assessed the regional patterns of warming and precipitation change for GMST 22 increase of 1.5°C and 2°C above 1850–1900 levels. Techniques used to represent the spatial variations in 23 climate at a given GWL are referred to as pattern scaling. 24 25 In the ‘traditional’ methodology as applied in the AR5 (Collins et al., 2013), patterns of climate change in 26 space are calculated as the product of the change in GSAT at a given point in time and a spatial pattern of 27 change that is constant over time and the scenario under consideration, and which may or may not depend on 28 a particular climate model (Allen and Ingram, 2002; Mitchell, 2003; Lambert and Allen, 2009; Andrews and 29 Forster, 2010; Bony et al., 2013; Lopez et al., 2014). This approach assumes that external forcing does not 30 affect the internal variability of the climate system, which may be regarded a stringent assumption when 31 taking into account decadal and multi-decadal variability (Lopez et al., 2014) and the potential nonlinearity 32 of the climate change signal. Moreover, pattern scaling is expected to have lower skill for variables with 33 large spatial variability (Tebaldi and Arblaster, 2014). Pattern scaling also fails to capture changes in 34 boundaries that moves poleward such as sea-ice extent and snow cover (Collins et al., 2013), and temporal 35 frequency quantities such as frost days that decrease under warming but are bounded at zero. Spatial patterns 36 are also expected to be different between transient and equilibrium simulations because of the long 37 adjustment timescale of the deep ocean. 38 39 Further developments of the AR5 approach have since explored the role of aerosols in modifying regional 40 climate responses at a specific degree of global warming and also the effect of different GCMs and scenarios 41 on the scaled spatial patterns (Frieler et al., 2012; Levy et al., 2013). Furthermore, the modified forcing- 42 response framework (Kamae and Watanabe, 2012, 2013; Sherwood et al., 2015), which decomposes the total 43 climate change into fast adjustments and slow response, identifies the fast adjustment as forcing-dependent 44 and the slow response as forcing-independent, scaling with the change in GSAT. 45 46 For precipitation change, there is near-zero fast adjustment for solar forcing but suppression during the fast- 47 adjustment phase for CO2 and black-carbon radiative forcing (Andrews et al., 2009; Bala et al., 2010; Cao et 48 al., 2015). By contrast, the slow response in precipitation change is independent of the forcing. This indicates 49 that pattern scaling is not expected to work well for climate variables that have a large fast-adjustment 50 component. Even in such cases, pattern scaling still works for the slow response component, but a correction 51 for the forcing-dependent fast adjustment would be necessary to apply pattern scaling to the total climate 52 change signal. In a multi-model setting, it has been shown that temperature change patterns conform better to 53 pattern scaling approximation than precipitation patterns (Tebaldi and Arblaster, 2014). 54 55 (Herger et al., 2015) have explored the use of multiple predictors for the spatial pattern of change at a given Do Not Cite, Quote or Distribute 4-17 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 degree of global warming, following the approach of Joshi et al. (2013) that explored the role of the land-sea 2 warming ratio as a second predictor. They found that the land-sea warming contrast changes in a non-linear 3 way with GSAT, and that it approximates the role of the rate of global warming in determining regional 4 patterns of climate change. The inclusion of the land-sea warming contrast as the second predictor provides 5 the largest improvement over the traditional technique. However, as pointed out by Herger et al. (2015), 6 multiple-predictor approaches still cannot detect nonlinearities (or internal variability), such as the apparent 7 dependence of spatial temperature variability in the mid- to high latitudes on GSAT (e.g. Screen, 2014; 8 Fischer and Knutti, 2014). 9 10 An alternative to the traditional pattern scaling approach is the time-shift method described by Herger et al. 11 (2015) which is applied in this chapter (also called the epoch approach, see Section 4.6.1). When applied to a 12 transient scenario such as SSP5-8.5, a future time-slab is referenced to a particular increase in the GSAT 13 (e.g., 1.5°C or 2°C of global warming above pre-industrial levels). The spatial patterns that result represent a 14 direct scaling of the spatial variations of climate change at the particular level of global warming. An 15 important advantage of this approach is that it ensures physical consistency between the different variables 16 for which changes are presented (Herger et al., 2015). The internal variability does not have to be scaled and 17 is consistent with the GSAT change. The time-shift method furthermore allows for a partial comparison of 18 how the rate of increase in GSAT influences the regional spatial patterns of climate change. For example, 19 spatial patterns of change for global warming of 2°C can be compared across the SSP2-4.5 and SSP5-8.5 20 scenarios. Direct comparisons can also be obtained between variations in the regional impacts of climate 21 change for the case where global warming stabilizes at, for instance, 1.5°C or 2°C of global warming, as 22 opposed to the case where the GSAT reaches and then exceeds the 1.5°C or 2°C thresholds (Tebaldi and 23 Knutti, 2018). An important potential caveat is that forcing mechanisms such as aerosol radiative forcing are 24 represented differently in different models, even for the same SSP. This may imply different regional aerosol 25 direct and indirect effects, implying different regional climate change patterns. Hence, it is important to 26 consider the variations in the forcing mechanisms responsible for a specific increase in GSAT towards 27 understanding the uncertainty range associated with the variations in regional climate change. A minor 28 practical limitation of this approach is that stabilization scenarios at 1.5°C or 2°C of global warming, such as 29 SSP1-2.6, do not allow for spatial patterns of change to be calculated from these scenarios at higher levels of 30 global warming, while it is possible in scenarios such as SSP5-8.5 (Herger et al., 2015). 31 32 In this chapter, the spatial patterns of change as a function of GWLs (defined in terms of the increase in the 33 GSAT relative to 1850–1900) are thus constructed using the time-shift approach, thereby accounting for 34 various nonlinearities and internal variability that influence the projected climate change signal. This implies 35 a reliance on large ensemble sizes to quantify the role of uncertainties in regional responses to different 36 degrees of global warming. The assessment in Section 4.6.1 also explores how the rate of global warming (as 37 represented by different SSPs), aerosol effects, and transient as opposed to stabilization scenarios influence 38 the spatial variations in climate change at specific levels of global warming. 39 40 41 4.2.5 Quantifying Various Sources of Uncertainty 42 43 The AR5 assessed with very high confidence that climate models reproduce the general features of the 44 global-scale annual mean surface temperature increase over the historical period, including the more rapid 45 warming in the second half of the 20th century, and the cooling immediately following large volcanic 46 eruptions. Furthermore, because climate and Earth system models are based on physical principles, 47 they were assessed in the AR5 to reproduce many important aspects of observed climate. Both aspects were 48 argued to contribute to our confidence in the models’ suitability for their application in quantitative future 49 predictions and projections (Flato et al., 2013). The AR6 assesses in Chapter 3, Section 3.8.2, with high 50 confidence that for most large-scale indicators of climate change, the recent mean climate simulated by the 51 latest generation climate models underpinning this assessment has improved compared to the models 52 assessed in the AR5, and with high confidence that the multi-model mean captures most aspects of observed 53 climate change well. These assessments form the foundation of applying climate and Earth system models to 54 the projections assessed in this chapter. Where appropriate, the assessment of projected changes is 55 accompanied by an assessment of process understanding and model evaluation. Do Not Cite, Quote or Distribute 4-18 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 That said, fitness-for-purpose of the climate models used for long-term projections is fundamentally difficult 3 to ascertain and remains an epistemological challenge (Parker, 2009; Frisch, 2015; Baumberger et al., 2017). 4 Some literature exists comparing previous IPCC projections to what has unfolded over the subsequent 5 decades (Cubasch et al., 2013), and recent work has confirmed that climate models since around 1970 have 6 projected global surface warming in reasonable agreement with observations once the difference between 7 assumed and actual forcing has been taken into account (Hausfather et al., 2020). However, the long-term 8 perspective to the end of the 21st century or even out to 2300 takes us beyond what can be observed in time 9 for a standard evaluation of model projections, and in this sense the assessment of long-term projections will 10 remain fundamentally limited. 11 12 The spread across individual runs within a multi-model ensemble represents the response to a combination of 13 different sources of uncertainties (see Chapter 1, Section 1.4.3), specifically: scenario uncertainties, climate 14 response uncertainties (also referred to as model uncertainties) related to parametric and other structural 15 uncertainties in the model representation of the climate system, and internal variability (e.g., Hawkins and 16 Sutton, 2009; Kirtman et al., 2013). While the nature of these uncertainties was introduced in Section 1.4.3, 17 this subsection assesses methods to disentangle different sources of uncertainties and quantify their 18 contributions to the overall ensemble spread. 19 20 As discussed extensively in the AR5 (Collins et al., 2013), ensemble spread in projections performed with 21 different climate models accounts for only part of the entire model uncertainty, even when considering the 22 uncertainty in the radiative forcing in projections (Vial et al., 2013) and forced response. The AR5 23 uncertainty characterisation (Kirtman et al., 2013) followed Hawkins and Sutton (2009) and diagnosed 24 internal variability through a high-pass temporal filter. This approach has deficiencies particularly if internal 25 variability manifests on the multi-decadal timescales (Deser et al., 2012a; Marotzke and Forster, 2015) and is 26 classified as (model) response uncertainty instead of internal variability. Single-model initial-condition large 27 ensembles revealed that the AR5 approach underestimates the role of internal variability uncertainty and 28 overestimates the role of model uncertainty (Maher et al., 2018; Stolpe et al., 2018; Lehner et al., 2020) 29 particularly at the local scale while yielding a reasonable approximation for uncertainty separation for GSAT 30 (Lehner et al., 2020). 31 32 Single-model initial-condition large ensembles thus represent a crucial step towards a cleaner separation of 33 model uncertainty and internal variability than available for the AR5 (Deser et al., 2014, 2016; Saffioti et al., 34 2017; Sippel et al., 2019; Milinski et al., 2020; von Trentini et al., 2020; Maher et al., 2021). Novel 35 approaches have been proposed to further quantify internal variability in multi-model ensembles (Hingray 36 and Saïd, 2014; Evin et al., 2019; Hingray et al., 2019). For time horizons beyond the limit of decadal 37 predictability (Branstator and Teng, 2010; Meehl et al., 2014; Marotzke et al., 2016), such as in the CMIP6 38 projections, the simulations are starting from random rather than assimilated initial conditions. Internal 39 variability constitutes an uncertainty in the projection of the climate in a future period of 10 or 20 years that 40 is irreducible, but can be precisely quantified for individual models using sufficiently large initial-condition 41 ensembles (Fischer et al., 2013; Deser et al., 2016; Hawkins et al., 2016; Pendergrass et al., 2017; Luo et al., 42 2018; Dai and Bloecker, 2019; Maher et al., 2019a; Deser et al., 2020). 43 44 Uncertainties in emissions of greenhouse gases and aerosols that affect future radiative forcings are 45 represented by selected SSP scenarios (Section 4.2.2, Section 1.6.1). In addition to emission uncertainties, 46 SSPs represent uncertainties in land use changes (van Vuuren et al., 2011)(Ciais et al., 2013)(O’Neill et al., 47 2016)(Christensen et al., 2018). Additional uncertainty comes from climate carbon-cycle feedbacks and the 48 residence time of atmospheric constituents, and are at least partly accounted for in emission-driven 49 simulations as opposed to concentration-driven simulations (Friedlingstein et al., 2014; Hewitt et al., 2016). 50 The climate carbon-cycle feedbacks affect the transient climate response to emissions (TCRE). Constraining 51 this uncertainty is crucial for the assessment of remaining carbon budgets consistent with global mean 52 temperature levels (Millar et al., 2017; Masson-Delmotte et al., 2018) and is covered in Chapter 5 of this 53 Report. Finally, there are uncertainties in future solar and volcanic forcing (see Cross-Chapter Box 4.1) 54 55 The relative magnitude of model uncertainty and internal variability depends on the time horizon of the Do Not Cite, Quote or Distribute 4-19 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 projection, location, spatial and temporal aggregation, variable, and signal strength (Rowell, 2012; Fischer et 2 al., 2013; Deser et al., 2014; Saffioti et al., 2017; Kirchmeier‐Young et al., 2019). New literature published 3 after the AR5 systematically discusses the role of different sources of uncertainty and shows that the relative 4 contribution of internal variability is larger for short than for long projection horizons (Marotzke and Forster, 5 2015; Lehner et al., 2020; Maher et al., 2021), larger for high latitudes than for low latitudes, larger for land 6 than for ocean variables, larger at station level than continental than global means, larger for annual 7 maxima/minima than for multi-decadal means, larger for dynamic quantities (and, by implication, 8 precipitation) than for temperature (Fischer et al., 2014). 9 10 The method introduced by Hawkins and Sutton (2009) and applied to GSAT projections reveals that by the 11 end of the 21st century, the fraction contribution of the climate model response uncertainty to the total 12 uncertainty is larger in CMIP6 than in CMIP5 whereas the relative contribution of scenario uncertainty is 13 smaller (Lehner et al., 2020). This is the case even when sub-selecting pathways and scenarios that are most 14 similar in CMIP5 and CMIP6, i.e. the range from RCP2.6 to RCP8.5 vs. SSP1-2.6 to SSP5-8-5, respectively 15 (Lehner et al., 2020). The larger range of response uncertainty is further consistent with the larger range of 16 TCR and GSAT warming for a comparable pathway in CMIP6 than CMIP5 (Forster et al., 2020; Tokarska et 17 al., 2020). 18 19 Some uncertainties are not, or only partially accounted for in the CMIP6 experiments, such as uncertainties 20 in natural forcings from solar and volcanic forcings, long-term Earth system feedbacks including land-ice 21 feedbacks, groundwater feedbacks (Smerdon, 2017) or some long-term carbon-cycle feedbacks (Fischer et 22 al., 2018). Where appropriate, this chapter uses results from non-CMIP ESMs or EMICs to assess the role of 23 these feedbacks. Still other uncertainties ⎼ such as further pandemics, nuclear holocaust, global natural 24 disaster such as tsunami or asteroid impact, or fundamental technological change such as fusion ⎼ are not 25 accounted for at all. 26 27 28 4.2.6 Display of Model Agreement and Spread 29 30 Maps of multi-model mean changes provide an average estimate for the forced model climate response to a 31 certain forcing. However, they do not include any information on the robustness of the response across 32 models nor on the significance of the change with respect to unforced internal variability (Tebaldi et al., 33 2011). Models can consistently show absence of significant change, in which case they should not be 34 expected to agree on the sign of a change (e.g., Tebaldi et al., 2011; Knutti and Sedláček, 2012; Fischer et 35 al., 2014). If a multi-model mean map of precipitation shows no change, it is unclear whether the models 36 consistently project insignificant changes or whether projections span both significant increases and 37 significant decreases. Several methods have been proposed to distinguish significant conflicting signals from 38 agreement on no significant change (Tebaldi et al., 2011; Knutti and Sedláček, 2013; McSweeney and Jones, 39 2013; Zappa et al., 2021). A set of different methods have been introduced in the literature to display model 40 robustness and to put a climate change signal into the context of internal variability. AR5 Box 12.1 provides 41 a detailed assessment of different methods of mapping model robustness and Cross-Chapter Box Atlas.1 42 provides an update of recent proposals including the methods used in this report. 43 44 Most methods for quantifying robustness assume that only one realization from each model is applied. There 45 are challenges that arise from having heterogeneous multi-model ensembles with many members for some 46 models and single members for others (Olonscheck and Notz, 2017; Evin et al., 2019). Furthermore, the 47 methods that map model robustness usually ignore that sharing parametrizations or entire components across 48 coupled models can lead to substantial model interdependence (Fischer et al., 2011; Kharin et al., 2012; 49 Knutti et al., 2013, 2017; Leduc et al., 2015; Sanderson et al., 2015, 2017; Annan and Hargreaves, 2017; 50 Boe, 2018; Abramowitz et al., 2019). This may lead to a biased estimate of model agreement if a substantial 51 fraction of models is interdependent. The methodologies and results in these literature since AR5 are higher 52 in quality and clarity. However, quantifying and accounting for model dependence in a robust way remains 53 challenging (Abramowitz et al., 2019). Furthermore, absence of significant mean change in a certain climate 54 variable does not imply absence of substantial impact, because there may be substantial change in variability, 55 which is typically not mapped (McSweeney and Jones, 2013). Do Not Cite, Quote or Distribute 4-20 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 Chapter 4 uses the advanced approach, taking into account the sign and significance of the change (Cross- 3 Chapter Box Atlas.1, approach C). Where not applicable, such as due to a lack of the necessary model 4 output, the simple method is used taking into account only agreement on the sign of the change across the 5 multi-model ensemble (Cross-Chapter Box Atlas.1, approach B). The advanced approach is similar to the 6 method used in AR5 but isolates conflicting signals as proposed in (Zappa et al., 2021). It uses three 7 mutually exclusive categories and distinguishes (a) areas with significant change and high model agreement 8 (no overlay), (b) areas with no change or no robust change (diagonal lines), and (c) areas with significant 9 change but low agreement (crossed lines). Category (a) marks areas where the climate change signals likely 10 emerge from internal variability, where two-thirds or more of the models project changes greater than 11 internal variability and 80% or more of the models agree on the sign of the change. Category (b) marks areas 12 where fewer than two-thirds of the models project changes greater than internal variability, and category (c) 13 marks areas with significant but conflicting signals, where two-thirds or more of the models project changes 14 greater than internal variability but less than 80% agree on the sign of the change. 15 16 In this chapter variability is defined as 1.645 ∙ √2 𝜎𝜎20𝑦𝑦𝑦𝑦 , where 𝜎𝜎20𝑦𝑦𝑦𝑦 is the standard deviation of 20-year 17 means in the pre-industrial control simulations (see Cross-Chapter Box Atlas.1). Category (a) uses a 18 definition very similar to the AR5 method for stippling except that the model signal is compared to its 19 corresponding internal rather than the multi-model mean variability, to account for the substantial model 20 differences in pre-industrial internal variability (Parsons et al., 2020). Changes smaller than internal 21 variability can have potential impacts particularly if they persist over sustained periods such as several 22 decades. Finally, even when changes do not exceed variability at the grid point level they may exceed 23 variability if aggregated over catchment basins, regions, or continents (Cross-Chapter Box Atlas.1). Maps of 24 mean changes also ignore potential changes in variability addressed by a more comprehensive assessment of 25 changes in temperature variability (Section 4.5.1) and modes of internal variability (Sections 4.4.3). 26 27 28 [START BOX 4.1 HERE] 29 30 BOX 4.1: Ensemble Evaluation and Weighting 31 32 AR5 used a pragmatic approach to quantify the uncertainty in CMIP5 GSAT projections (Collins et al., 33 2013). The multi-model ensemble was constructed by picking one realization per model per scenario. For 34 most quantities, the 5–95% ensemble range was used to characterize the uncertainty, but the 5–95% 35 ensemble range was interpreted as the 17–83% (likely) uncertainty range. The uncertainty was thus explicitly 36 assumed to contain sources not represented by the model range. While straightforward and clearly 37 communicated, this approach had several drawbacks. 38 i) The uncertainty breakdown into scenario uncertainty, model uncertainty, and internal variability 39 (Cox and Stephenson, 2007; Hawkins and Sutton, 2009) in the AR5 followed Hawkins and Sutton 40 (2009) and diagnosed internal variability through a high-pass temporal filter (Kirtman et al., 2013), but 41 it has since become clear that even multi-decadal trends contain substantial internal variability relative 42 to the forced response in many variables (e.g., (Deser et al., 2012a; Marotzke and Forster, 2015; Deser 43 et al., 2020; Lehner et al., 2020)); hence a more comprehensive approach is needed. 44 ii) The uncertainty characterization ignores observation-based information about internal climate 45 variability during the most recent past, such as is used in initialized predictions. While this may matter 46 little for the long-term projections (Collins et al., 2013), it is very important for the near-term future 47 (Kirtman et al., 2013). AR5 included additional uncertainty quantification for the near-term 48 projections (Kirtman et al., 2013), leading to a downward adjustment of assessed near-term GSAT 49 change, which created an inconsistency in the transition from near-term to long-term GSAT 50 assessment in the AR5. 51 iii) AR5 used the range of CMIP5 equilibrium climate sensitivity (ECS) side-by-side with the 52 ECS likely range assessed from multiple lines of evidence (the CMIP5 ensemble, instrumental 53 observations, and paleo-information, (Collins et al., 2013)). While the CMIP5 range in ECS and the 54 AR5 ECS likely range did not differ much, the difference did create an inconsistency. Furthermore, 55 AR5 WGIII used the assessed likely range for ECS in their calculations of carbon budgets (IPCC, Do Not Cite, Quote or Distribute 4-21 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2014), and these uncertainties matter a great deal when assessing remaining carbon budgets consistent 2 with limiting global warming to 1.5°C above pre-industrial levels (Millar et al., 2017, 2018a, 2018b; 3 Schurer et al., 2018)(Rogelj et al., 2018b). 4 5 Another important consideration concerns the potential weighting of model contributions to an ensemble, 6 based on model independence, model performance during the historical period, or both. Such model 7 weighting (in fact, model selection) was performed in the AR5 for projections of Arctic sea ice (Collins et 8 al., 2013), but that particular application has subsequently been shown by Notz (2015) to be contaminated by 9 internal variability, making the resulting weighting questionable (see also Stroeve and Notz (2015)). For a 10 general cautionary note, see Weigel et al. (2010). Approaches that take into account internal variability and 11 model independence have been proposed since AR5 (Knutti et al., 2017; Boe, 2018; Abramowitz et al., 12 2019; Brunner et al., 2020). 13 14 There are hence good reasons for basing an assessment of future global climate on lines of evidence in 15 addition to the projection simulations. However, despite some progress, no universal, robust method for 16 weighting a multi-model projection ensemble is available, and expert judgement must be included, as it did 17 for AR5, in the assessment of the projections. The default in this chapter follows the AR5 approach for 18 GSAT (Collins et al., 2013) and interprets the CMIP6 5–95% ensemble range as the likely uncertainty range. 19 20 Additional lines of evidence enter the assessment particularly for the most important indicator of global 21 climate change, GSAT. The CMIP6 ensemble generally shows larger projected warming by the end of the 22 21st century, relative to the average over the period 1995–2014, than the CMIP5 ensemble (Section 4.3.1). 23 The warming has increased in part because of models with higher ECS in CMIP6, compared to CMIP5 (e.g., 24 Meehl et al., 2020; Tokarska et al., 2020; Zelinka et al., 2020; Zhu et al., 2020, high confidence), and in part 25 because of higher ERF in CMIP6 than in CMIP5 (e.g., Tebaldi et al., 2021, Section 4.6.2). Because change 26 in several other important climate quantities scales with change in GSAT (Section 4.2.4), bringing in 27 additional lines of evidence is particularly important for the GSAT assessment. 28 29 The Chapter 4 assessment uses information from the following sources: 30 31 (i) The CMIP6 multi-model ensemble (Eyring et al., 2016), augmented if appropriate by the CMIP5 32 ensemble (Taylor et al., 2012). 33 (ii) Single-model large initial-condition ensembles (e.g., Kay et al., 2015; Sigmond and Fyfe, 2016; 34 Maher et al., 2019) and combinations of control runs with CMIP transient simulations (e.g., 35 Thompson et al., 2015; Olonscheck and Notz, 2017) to characterize internal variability. Several 36 analyses using multiple large ensembles have recently become available and add robustness to 37 the results (Maher et al., 2018, 2019b, 2020; Deser et al., 2020; Lehner et al., 2020)(Maher et al., 38 2021). 39 (iii) Assessed best estimates, likely, and very likely ranges of ECS and TCR, from process 40 understanding, warming in the instrumental record, paleoclimates, and emergent constraints 41 (Chapter 7, Table 7.13, Table 7.14, Section 7.5). The ECS and TCR ranges are converted into 42 GSAT ranges using as an emulator a two-layer energy balance model (EBM, e.g., Held et al., 43 2010) that is driven by the effective radiative forcing (ERF) assessed in Chapter 7 (see Cross- 44 Chapter Box 7.1). Assuming for the ERF resulting from a doubling of the CO2 concentration, 45 ∆F2 • CO2=4.0 W m-2 (close to the best estimate of 3.93 W m-2, Section 7.3), and using the so- 46 called zero-layer approximation to the EBM (e.g., Marotzke and Forster, 2015; Jiménez-de-la- 47 Cuesta and Mauritsen, 2019) permits a one-to-one translation of any pair of ECS and TCR into a 48 pair of climate feedback parameter α and ocean heat uptake coefficient κε, using the simple 49 equations α=−∆F2 • CO2 ECS-1 and κε= ∆F2 • CO2 TCR-1 − ∆F2 • CO2 ECS-1 (e.g., Jiménez-de-la- 50 Cuesta and Mauritsen, 2019); see Chapter 7 for a detailed discussion). The results are displayed 51 in Box 4.1, Figure 1 and are used in the synthesis GSAT assessment in Section 4.3.4. 52 (iv) Model independence diagnosed a priori, based on shared model components for atmosphere, 53 ocean, land surface, and sea ice of CMIP5 models (Boe, 2018). CMIP5 models have been re- 54 sampled assuming that two models sharing either the atmosphere or the ocean component are 55 effectively the same model (Maher et al., 2021). Downweighting CMIP5 models that share a Do Not Cite, Quote or Distribute 4-22 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 component with another has substantial influence on diagnosed model agreement on change in 2 ENSO (Maher et al., 2021), but has negligible influence (much less than 0.1°C) on the ensemble 3 mean and range of GSAT change over the 21st century. No corresponding diagnosis exists yet 4 for CMIP6 models, and no weighting based on a-priori independence is applied here. 5 (v) Performance in simulating the past and a-posteriori independence based on comparison against 6 observations (Knutti et al., 2017; Abramowitz et al., 2019). This approach has been applied to 7 CMIP6-simulated GSAT and has led to a substantial reduction in model range (Liang et al., 8 2020, (Brunner et al., 2020), Section 4.3.4). CMIP6-simulated Arctic sea-ice area has been 9 compared to the observed record, and models have been selected whose ensemble range across 10 their individual realizations (Olonscheck and Notz, 2017) includes the observational range of 11 uncertainty. A larger fraction of these selected simulations show an ice-free Arctic in September 12 before 2050, compared to the entire CMIP6 ensemble (Notz and SIMIPCommunity, 2020, 13 Section 4.3.2). 14 (vi) A linear inverse method (kriging) has combined the entire GSAT record since 1850 with the 15 CMIP6 historical simulations to produce constrained projections for the 21st century; again the 16 reduction in range has been substantial (Ribes et al., 2021; Section 4.3.4). 17 (vii) Emergent constraints (e.g., (Hall and Qu, 2006; Cox et al., 2018; Brient, 2020), which for the 18 post-1970 warming have been applied to the CMIP5 (Jiménez-de-la-Cuesta and Mauritsen, 19 2019) and CMIP6 ensembles (Nijsse et al., 2020; Tokarska et al., 2020) and have likewise led to 20 a substantial reduction in GSAT ensemble range (Section 4.3.4). 21 (viii) Climate predictions initialized from recent observations (e.g., Kirtman et al., 2013) and the 22 Decadal Climate Prediction Project (DCPP) contribution to CMIP6 (Boer et al., 2016; Smith et 23 al., 2020; Sospedra-Alfonso and Boer, 2020). Initialized predictions for the period 2019–2028 24 exist for eight DCPP models and are used here (Box 4.1 Figure 1, Section 4.4.1). The DCPP 25 results have been drift-removed and referenced to the time-averaged hindcasts for 1995–2014 26 lead-year by lead-year, following (Kharin et al., 2012; Kruschke et al., 2016). 27 28 Box 4.1 Figure 1 shows annual-mean GSAT simulated by CMIP6 models for both the historical period and 29 forced by scenario SSP2-4.5 until 2100, combined with various characterizations of uncertainty. First, 30 internal variability is estimated with the 50-member ensemble simulated with CanESM5. The 5–95% 31 ensemble range for annual-mean GSAT in CanESM5 is slightly below 0.4°C; in other CMIP6 large 32 ensembles this range is about 0.5°C (MIROC6, IPSL-CM6A) and slightly above 0.6°C (S-LENS/EC- 33 Earth3). The CMIP5 large ensemble MPI-GE shows a range of slightly below 0.5°C (Bengtsson and Hodges, 34 2019), in reasonable agreement with observed variability (Maher et al., 2019b). There is thus high 35 confidence in the CMIP6-simulated level of internal variability in annual-mean GSAT, as displayed in Box 36 4.1, Figure 1. 37 38 Second, Section 7.5 very likely ECS and TCR ranges are converted into GSAT ranges with the EBM as an 39 emulator using, in this example, SSP2-4.5 radiative forcing information. Because the ECS and TCR 40 assessments in Section 7.5 are based on multiple lines of evidence and the EBM physics are well understood, 41 there is likewise high confidence in the EBM-emulated warming. Third, the initialized-forecast ensembles 42 from eight CMIP6 DCPP models are shown in the inset, for the period 2019–2028. During this period, the 43 initialized forecasts are consistent, within internal variability, with the EBM-emulated range, further adding 44 to the high confidence in the assessed-GSAT range. 45 46 The constrained range of GSAT change is useful for quantifying uncertainties in changes of other climate 47 quanties that scale well with GSAT change, such as September Arctic sea-ice area, global-mean 48 precipitation, and many climate extremes (Cross-Chapter Box 11.1). However, there are also quantities that 49 do not scale linearly with GSAT change, such as global-mean land precipitation, atmospheric circulation, 50 AMOC, and modes of variability, especially ENSO SST variability. Because we do not have robust scientific 51 evidence to constrain changes in other quantities, uncertainty quantification for their changes is based on 52 CMIP6 projections and expert judgement. For the assessment for changes in GMSL, the contribution from 53 land-ice melt has been added offline to the CMIP6 simulated contributions from thermal expansion, 54 consistent with Chapter 9 (see Section 9.6). 55 Do Not Cite, Quote or Distribute 4-23 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 [START BOX 4.1, FIGURE 1 HERE] 3 4 Box 4.1 Figure 1: CMIP6 annual-mean GSAT simulations and various contributions to uncertainty in the 5 projections ensemble. The figure shows anomalies relative to the period 1995–2014 (left y-axis), 6 converted to anomalies relative to 1850–1900 (right y-axis); the difference between the y-axes is 7 0.85°C (Cross-Chapter Box 2.3). Shown are historical simulations with 39 CMIP6 models (grey) 8 and projections following scenario SSP2-4.5 (dark yellow; thin lines: individual simulations; 9 heavy line; ensemble mean; dashed lines: 5% and 95% ranges). The black curve shows the 10 observations-based estimate (HadCRUT5, (Morice et al., 2021)). Light blue shading shows the 50- 11 member ensemble CanESM5, such that the deviations from the CanESM5 ensemble mean have 12 been added to the CMIP6 multi-model mean. The green curves are from the emulator and show 13 the central estimate (solid) and very likely range (dashed) for GSAT. The inset shows a cut-out 14 from the main plot and additionally in light purple for the period 2019–2028 the initialized 15 forecasts from eight models contributing to DCPP (Boer et al., 2016); the deep-purple curve shows 16 the average of the forecasts. Further details on data sources and processing are available in the 17 chapter data table (Table 4.SM.1). 18 19 [END BOX 4.1, FIGURE 1 HERE] 20 21 22 [END BOX 4.1 HERE] 23 24 25 4.3 Projected Changes in Global Climate Indices in the 21st Century 26 27 This section assesses the latest simulations of representative indicators of global climate change presented as 28 time series and tabulated values over the 21st century and across the main realms of the global climate 29 system. In the atmospheric realm (see Section 4.3.1), we assess simulations of GSAT (see Figure 4.2a) and 30 global land precipitation (see Figure 4.2b). Across the cryospheric, oceanic, and biospheric realms (see 31 Section 4.3.2), we assess simulations of Arctic SIA (see Figure 4.2c), GMSL (see Figure 4.2d), the AMOC, 32 ocean and land carbon uptake, and pH. In Section 4.3.3 we assess simulations of several indices of climate 33 variability, namely, the indices of the NAM, SAM, and ENSO. Finally, Section 4.3.4 assesses future GSAT 34 change based on the CMIP6 ensemble in combination with other lines of evidence. An assessment of 35 projected changes in related global extreme indices can be found in Chapter 11. 36 37 38 [START FIGURE 4.2 HERE] 39 40 Figure 4.2: Selected indicators of global climate change from CMIP6 historical and scenario simulations. (a) 41 Global surface air temperature changes relative to the 1995–2014 average (left axis) and relative to the 42 1850–1900 average (right axis; offset by 0.82°C, which is the multi-model mean and close to observed 43 best estimate, Cross-Chapter Box 2.1, Table 1). (b) Global land precipitation changes relative to the 44 1995–2014 average. (c) September Arctic sea-ice area. (d) Global mean sea-level change (GMSL) 45 relative to the 1995–2014 average. (a), (b) and (d) are annual averages, (c) are September averages. In 46 (a)-(c), the curves show averages over the CMIP6 simulations, the shadings around the SSP1-2.6 and 47 SSP3-7.0 curves show 5–95% ranges, and the numbers near the top show the number of model 48 simulations used. Results are derived from concentration-driven simulations. In (d), the barystatic 49 contribution to GMSL (i.e., the contribution from land-ice melt) has been added offline to the CMIP6 50 simulated contributions from thermal expansion (thermosteric). The shadings around the SSP1-2.6 and 51 SSP3-7.0 curves show 5–95% ranges. The dashed curve is the low confidence and low likelihood 52 outcome at the high end of SSP5-8.5 and reflects deep uncertainties arising from potential ice-sheet and 53 ice-cliff instabilities. This curve at year 2100 indicates 1.7 m of GMSL rise relative to 1995–2014. More 54 information on the calculation of GMSL are available in Chapter 9, and further regional details are 55 provided in the Atlas. Further details on data sources and processing are available in the chapter data table 56 (Table 4.SM.1). 57 58 [END FIGURE 4.2 HERE] Do Not Cite, Quote or Distribute 4-24 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 3 From the CMIP6 multi-model ensemble we consider historical simulations with observed external forcings 4 to 2014 and extensions to 2100 based on the five high-priority scenarios. We use the first realization (‘r1’) 5 contributed by each modelling group. In tabular form, we show ensemble-mean changes and uncertainties 6 for the near-term (2021–2040), mid-term (2041–2060), and the long-term (2081–2100), relative to present- 7 day (1995–2014) and the approximation to pre-industrial (1850–1900). Changes in precipitation over land 8 near 1.5°C, 2.0°C, 3.0°C, and 4.0°C of global warming relative to 1850–1900 are also assessed. 9 10 11 4.3.1 Atmosphere 12 13 4.3.1.1 Surface Air Temperature 14 15 The AR5 assessed from CMIP5 simulations and other lines of evidence that GSAT will continue to rise over 16 the 21st century if GHG concentrations continue increasing (Collins et al., 2013). The AR5 concluded that 17 GSAT for 2081–2100, relative to 1986–2005 will likely be in the 5–95% range of 0.3°C–1.7°C under 18 RCP2.6 and 2.6°C–4.8°C under RCP8.5. The corresponding ranges for the intermediate emission scenarios 19 with emissions peaking around 2040 (RCP4.5) and 2060 (RCP6.0) are 1.1°C–2.6°C and 1.4°C–3.1°C, 20 respectively. The AR5 further assessed that GSAT averaged over the period 2081–2100 are projected to 21 likely exceed 1.5°C above 1850–1900 for RCP4.5, RCP6.0 and RCP8.5 (high confidence) and are likely to 22 exceed 2°C above 1850–1900 for RCP6.0 and RCP8.5 (high confidence). Global surface temperature 23 changes above 2°C under RCP2.6 were deemed unlikely (medium confidence). 24 25 Here, for continuity’s sake, we assess the CMIP6 simulations of GSAT in a fashion similar to the AR5 26 assessment of the CMIP5 simulations. From these, we compute anomalies relative to 1995–2014 and display 27 the evolution of ensemble means and 5–95% ranges (see Figure 4.2). We also use the ensemble mean GSAT 28 difference between 1850–1900 and 1995–2014, 0.82°C, to provide an estimate of the changes since 1850– 29 1900 (see the right axis on Figure 4.2). Finally, we tabulate the ensemble mean changes between 1995–2014 30 and 2021–2040, 2041–2060, and 2081–2100 respectively (see Figure 4.2). 31 32 The CMIP6 models show a 5–95% range of GSAT change for 2081–2100, relative to 1995–2014, of 0.6°C– 33 2.0°C under SSP1-2.6 where CO2 concentrations peak between 2040 and 2060 (see Table 4.2). The 34 corresponding range under the highest overall emission scenario (SSP5-8.5) is 2.7°C–5.7°C. The ranges for 35 the intermediate emission scenarios (SSP2-4.5 and SSP3-7.0), where CO2 concentrations increase to 2100, 36 but less rapidly than SSP5-8.5, are 1.4°C–3.0°C and 2.2°C–4.7°C, respectively. The range for the lowest 37 emission scenario (SSP1-1.9) is 0.2°C–1.3°C. 38 39 In summary, the CMIP6 models show a general tendency toward larger long-term globally averaged surface 40 warming than did the CMIP5 models, for nominally comparable scenarios (very high confidence). In SSP1- 41 2.6 and SSP2-4.5, the 5–95% ranges have remained similar to the ranges in RCP2.6 and RCP4.5, 42 respectively, but the distributions have shifted upward by about 0.3°C (high confidence). For SSP5-8.5 43 compared to RCP8.5, the 5% bound of the distribution has hardly changed, but the 95% bound and the range 44 have increased by about 20% and 40%, respectively (high confidence). About half of the warming increase 45 has occurred because of more models with higher climate sensitivity in CMIP6, compared to CMIP5; the 46 other half of the warming increase arises from higher effective radiative forcing in nominally comparable 47 scenarios (medium confidence, see Section 4.6.2). 48 49 50 [START TABLE 4.2 HERE] 51 52 Table 4.2: CMIP6 annual mean surface air temperature anomalies (°C). Displayed are multi-model averages 53 and, in parentheses, the 5–95% ranges, for selected time periods, regions, and SSPs. The numbers of 54 models used are indicated in Figure 4.2. 55 Do Not Cite, Quote or Distribute 4-25 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI Units = °C SSP1-1.9 SSP1-2.6 SSP2-4.5 SSP3-7.0 SSP5-8.5 Global: 2021–2040 relative to 1995–2014 0.7 (0.3, 1.1) 0.7 (0.4, 1.1) 0.7 (0.4, 1.2) 0.7 (0.5, 1.2) 0.8 (0.5, 1.3) relative to 1850–1900 1.5 (1.1, 2.2) 1.6 (1.1, 2.2) 1.6 (1.0, 2.3) 1.6 (1.0, 2.4) 1.7 (1.2, 2.4) Global: 2041–2060 relative to 1995–2014 0.8 (0.3, 1.5) 1.0 (0.6, 1.6) 1.3 (0.8, 1.9) 1.4 (0.9, 2.3) 1.7 (1.2, 2.5) relative to 1850–1900 1.7 (1.1, 2.4) 1.9 (1.2, 2.7) 2.1 (1.5, 3.0) 2.3 (1.6, 3.2) 2.6 (1.8, 3.4) Global: 2081–2100 relative to 1995–2014 0.7 (0.2, 1.5) 1.2 (0.6, 2.0) 2.0 (1.4, 3.0) 3.1 (2.2, 4.7) 4.0 (2.7, 5.7) relative to 1850–1900 1.5 (1.0, 2.2) 2.0 (1.3, 2.8) 2.9 (2.1, 4.0) 3.9 (2.8, 5.5) 4.8 (3.6, 6.5) Land: 2081–2100 0.9 (0.3, 2.0) 1.5 (0.8, 2.6) 2.7 (1.7, 4.0) 4.1 (3.0, 6.2) 5.3 (3.5, 7.6) relative to 1995–2014 Ocean: 2081–2100 0.6 (0.1, 1.2) 1.0 (0.5, 1.8) 1.8 (1.2, 2.7) 2.7 (1.8, 4.0) 3.4 (2.3, 4.9) relative to 1995–2014 Tropics: 2081–2100 0.5 (0.1, 1.1) 1.0 (0.5, 1.6) 1.8 (1.2, 2.5) 2.7 (2.0, 4.0) 3.5 (2.4, 4.9) relative to 1995–2014 Arctic: 2081–2100 2.4 (0.5, 6.6) 3.3 (0.4, 7.5) 5.4 (2.8, 10.0) 7.7 (4.5, 13.4) 10.0 (6.2, 15.2) relative to 1995–2014 Antarctic: 2081–2100 0.5 (0.0, 1.1) 1.1 (0.1, 2.9) 1.9 (0.6, 3.2) 2.8 (1.3, 4.5) 3.6 (1.7, 5.6) relative to 1995–2014 1 2 [END TABLE 4.2 HERE] 3 4 5 With regards to global warming levels (GWL) of 1.5°C, 2.0°C, and 3.0°C, we note that there is unanimity 6 across all of the CMIP6 model simulations that GSAT change relative to 1850–1900 will rise above: 1) 7 1.5°C following SSP2-4.5, SSP3-7.0, or SSP5-8.5 (on average around 2030); 2) 2.0°C following either 8 SSP3-7.0 or SSP5-8.5 (on average around 2043), and 3) 3.0°C following SSP5-8.5 (on average around 9 2062). Under SSP1-1.9, 55% and 36% of the model simulations rise above 1.5°C and 2.0°C, respectively, 10 while for SSP1-2.6 those percentages increase to 87% and 58%, respectively. Here, the time of GSAT 11 exceedance is determined as the first year at which 21-year running averages of GSAT exceed the given 12 GWL. In Section 4.3.4, these values are reassessed using CMIP6 ensemble in combination with other lines 13 of evidence. 14 15 CMIP6 models project increases in area-weighted land, ocean, tropical (30°S–30°N), Arctic (67.7°N–90°N), 16 and Antarctic (90°S–55°S) surface air temperature (see Table 4.2). Consistent with AR5, and earlier 17 assessments, CMIP6 models project that annual average surface air temperature will warm about 50% more 18 over land than over the ocean, and that the Arctic will warm about more than 2.5 times the global average 19 (see Section 4.5.1). For 2081–2100, relative to 1995–2014, the CMIP6 models show 5–95% ranges of 20 warming over land of 0.3°C–2.0°C and 3.5°C–7.6°C following SSP1-1.9 and SSP5-8.5, respectively. The 21 corresponding ranges for Arctic surface air temperature change are 0.5°C–6.6°C and 6.2°C–15.2°C, 22 respectively. 23 24 The concentration-driven simulations presented above use a prescribed CO2 pathway calculated by the 25 MAGICC7.0 model using the CMIP6 emissions (Meinshausen et al., 2020). This is compared here with the 26 CO2 concentration simulated by CMIP6 ESMs in response to the SSP5-8.5 emissions (Figure 4.3). The 27 1995–2014 mean simulated CO2 level is 375 ppm, very similar to the prescribed 378 ppm, but the ESM 5– 28 95% range is 357–391 ppm. By the end of the 21st century (2081–2100), the ESM mean is 953 ppm ⎼ below 29 the prescribed CO2 pathway (1004 ppm), but with a large 5–95% range of 848–1045 ppm, which spans the 30 prescribed concentration level. This result differs from CMIP5, which showed that ESMs typically simulated 31 CO2 concentrations higher than the prescribed concentration-driven RCP pathways. Reduced spread in 32 CMIP6 carbon cycle feedbacks compared to CMIP5 has been postulated to be due to the inclusion of 33 nitrogen cycle processes in about half of CMIP6 ESMs (Arora et al., 2020). This means that the CMIP6 34 spread in GSAT response to CO2 emissions is dominated by climate sensitivity differences between ESMs 35 more than by carbon cycle differences (Jones and Friedlingstein, 2020; Williams et al., 2020) (high 36 confidence). 37 38 Simulated GSAT over 1995–2014, relative to 1850–1900 period, warms by very similar amounts in the two Do Not Cite, Quote or Distribute 4-26 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 sets of simulations: 0.82°C (0.45–1.31) in emissions-driven compared with 0.75°C (0.53–1.09) in 2 concentration-driven simulations. By the end of the 21st century, warming in emissions-driven simulations is 3 very similar: 4.58°C (3.53–6.70), reflecting the slightly lower CO2 concentration simulated by the ESMs 4 compared with warming under the prescribed CO2 pathway of 4.69°C (3.70–6.77). This difference in model- 5 mean response is more than an order of magnitude smaller than the 5–95% spread across model projections. 6 The spread in CO2 concentration, compared with the prescribed default concentration, leads to a very small 7 increase by about 0.1°C in the spread of GSAT projections, but it is not possible to tell if this is a direct 8 consequence of the simulation configuration or internal variability of the model simulations. These 9 differences due to experimental configuration would be smaller still under scenarios with lower CO2 levels, 10 and so we assess that results from concentration-driven and emissions-driven configurations do not affect the 11 assessment of GSAT projections (high confidence). 12 13 14 [START FIGURE 4.3 HERE] 15 16 Figure 4.3: Comparison of concentration-driven and emission-driven simulation. (a) Atmospheric CO2 17 concentration, (b) GSAT from models which performed SSP5-8.5 scenario simulations in both emissions- 18 driven (blue; esm-ssp585) and concentration-driven (red; ssp585) configurations. For concentration 19 driven simulations, CO2 concentration is prescribed, and follows the red line in panel (a) in all models. 20 For emissions-driven simulations, CO2 concentration is simulated and can therefore differ for each model, 21 blue lines in panel (a). Further details on data sources and processing are available in the chapter data 22 table (Table 4.SM.1). 23 24 [END FIGURE 4.3 HERE] 25 26 27 4.3.1.2 Precipitation 28 29 AR5 assessed from CMIP5 projections that global mean precipitation over the 21st century will increase by 30 more than 0.05 mm day–1 (about 2% of global precipitation) and 0.15 mm day–1 (about 5% of global 31 precipitation) under the RCP2.6 and RCP8.5 scenarios, respectively (Collins et al., 2013). These changes are 32 generally in line with those from the CMIP6 simulations following SSP1-2.6 and SSP5-8.5 (see Table 4.3). 33 34 Unlike AR5, our focus here is on land rather than global precipitation because land precipitation has greater 35 societal relevance. These are displayed as percent changes relative to 1995–2014 (see Figure 4.2b). Based 36 on these results, we conclude that global land precipitation is larger during the period 2081–2100 than during 37 the period 1995–2014, under all scenarios considered here (see Table 4.3: high confidence). Global land 38 precipitation for 2081–2100, relative to 1995–2014, shows a 5–95% range of –0.2–4.7% under SSP1-1.9 and 39 0.9–12.9% under SSP5-8.5, respectively. The corresponding ranges under the other emission scenarios are 40 0.0–6.6% (SSP1-2.6), 1.5–8.3% (SSP2-4.5), and 0.5–9.6% (SSP3-7.0). A detailed assessment of 41 hydrological sensitivity, or change in precipitation per degree warming, can be found in Chapter 8, Section 42 8.2.1. 43 44 For scenarios where unanimity across all of the model simulations that GSAT change relative to 1850–1900 45 rises above 1.5°C (SSP2-4.5, SSP3-7.0, or SSP5-8.5), the ensemble-mean change in global land precipitation 46 from 1850–1900 until the time of exceedance is on average about 1.6%. For scenarios with unanimous 47 global warming above 2.0°C (SSP3-7.0, or SSP5-8.5) and 3.0°C (SSP5-8.5), the ensemble-mean increase in 48 global land precipitation for those models that do exceed 2.0°C and 3.0°C is on average about 2.6% and 49 4.9%, respectively. On average under SSP1-1.9 and SSP1-2.6, the global land precipitation change for 50 simulations where global warming exceeds 1.5°C and 2.0°C will be about 1.9% and 3.0%, respectively. 51 52 53 [START TABLE 4.3 HERE] 54 55 Table 4.3: CMIP6 precipitation anomalies (%) relative to averages over 1995–2014 for selected future periods, 56 regions and SSPs. Displayed are the multi-model averages across the individual models and, in Do Not Cite, Quote or Distribute 4-27 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 parentheses, the 5–95% ranges. Also shown are land precipitation anomalies at the time when global 2 increase in GSAT relative to 1850–1900 exceeds 1.5°C, 2.0°C, 3.0°C, and 4.0°C, and the percentage of 3 simulations for which such exceedances are true (to the right of the parentheses). Here, the time of GSAT 4 exceedance is determined as the first year at which 21-year running averages of GSAT exceed the given 5 threshold. Land precipitation percent anomalies are then computed as 21-year averages about the year of 6 the first GSAT crossing. The numbers of models used are indicated in Figure 4.4. 7 Units = % SSP1-1.9 SSP1-2.6 SSP2-4.5 SSP3-7.0 SSP5-8.5 Land: 2021–2040 2.4 (0.7, 4.1) 2.0 (–0.6, 3.6) 1.5 (–0.4, 3.6) 1.2 (–1.0, 3.4) 1.7 (–0.1, 4.1) 2041–2060 2.7 (0.6, 5.0) 2.8 (–0.4, 5.2) 2.7 (0.3, 5.2) 2.5 (–0.8, 5.1) 3.7 (–0.1, 6.9) 2081–2100 2.4 (–0.2, 4.7) 3.3 (0.0, 6.6) 4.6 (1.5, 8.3) 5.8 (0.5, 9.6) 8.3 (0.9, 12.9) Global: 2081–2100 2.0 (0.4, 4.2) 2.9 (1.0, 5.2) 4.0 (2.3, 6.7) 4.7 (2.3, 8.2) 6.5 (3.4, 10.9) Ocean: 2081–2100 1.9 (0.6, 4.1) 2.8 (1.1, 5.4) 3.8 (2.0, 6.8) 4.4 (2.1, 7.9) 6.0 (2.9, 10.5) Land: ∆T > 1.5°C 2.0 (0.6, 4.4) 55 1.7 (–2.0, 6.9) 87 1.7 (–2.9, 6.2) 100 1.5 (–3.9, 6.6) 100 1.5 (–3.5, 6.4) 100 ∆T > 2.0°C 3.8 (2.4, 5.8) 36 2.2 (–2.0, 4.6) 58 2.8 (–2.2, 8.1) 97 2.4 (–4.4, 7.7) 100 2.8 (–2.8, 8.3) 100 ∆T > 3.0°C - (-, -) 0 - (-, -) 0 4.9 (1.5, 9.6) 54 4.3 (–4.4, 11.5) 97 4.9 (–2.6, 11.0) 100 ∆T > 4.0°C - (-, -) 0 - (-, -) 0 4.2 (1.3, 6.3) 9 5.1 (–2.5, 11.1) 57 6.4 (–3.4, 15.0) 85 8 9 [END TABLE 4.3 HERE] 10 11 12 Relative to 1995–2014, and across all of the scenarios considered here, CMIP6 models show greater 13 increases in precipitation over land than either globally or over the ocean (see Table 4.3; high confidence). 14 Over the Northern Hemisphere Extratropics, the 5–95% changes in precipitation over land between 1995– 15 2014 and 2021–2040, 2041–2060, and 2081–2100, following SSP5-8.5, are 0.6–4.9%, 1.5–8.8%, and 4.7– 16 17.2%, respectively (Figure 4.4). At the other end of scenario spectrum, SSP1-1.9, the corresponding 17 changes are 0.6–5.4%, 0.6–7.3%, and 0.2–7.7%, respectively. By contrast, over the North Atlantic 18 subtropics, precipitation decreases by about 10% following SSP3-7.0 and SSP5-8. There is no change in 19 subtropical precipitation in the North Atlantic following SSP1-1.9, SSP1-2.6, or SSP2-4.5 (high confidence); 20 thereby highlighting the potential limitations of pattern scaling for regional hydrological changes (see also 21 Section 8.5.3). The reasons for the opposing changes in these two regions are assessed in Chapter 8. 22 23 24 [START FIGURE 4.4 HERE] 25 26 Figure 4.4: CMIP6 annual mean precipitation changes (%) from historical and scenario simulations. (a) 27 Northern Hemisphere extratropics (30°N–90°N). (b) North Atlantic subtropics (5°N–30°N, 80°W–0°). 28 Changes are relative to 1995–2014 averages. Displayed are multi-model averages and, in parentheses, 5– 29 95% ranges. The numbers inside each panel are the number of model simulations. Results are derived 30 from concentration-driven simulations. Further details on data sources and processing are available in the 31 chapter data table (Table 4.SM.1). 32 33 [END FIGURE 4.4 HERE] 34 35 36 4.3.2 Cryosphere, Ocean, and Biosphere 37 38 4.3.2.1 Arctic Sea Ice 39 40 AR5 assessed from CMIP5 simulations that there will be year-round reductions of Arctic sea ice coverage by 41 the end of this century (Collins et al., 2013). These range from 43% under RCP2.6 and 94% under RCP8.5 in 42 September, and from 8% under RCP2.6 and 34% under RCP8.5 in March (medium confidence). Based on a Do Not Cite, Quote or Distribute 4-28 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 five-member selection of CMIP5 models, AR5 further assessed that for RCP8.5, Arctic sea-ice coverage in 2 September will drop below 1 million km2 and be practically ice free at some point between 2040 and 2060. 3 SROCC further assessed that the probability of an ice-free Arctic in September for stabilized global warming 4 of 1.5°C and 2.0°C is approximately 1% and 10–35%, respectively (IPCC, 2019). 5 6 With regards to the model selection in AR5, model evaluation studies have since identified shortcomings of 7 the CMIP5 models to match the observed distribution of sea-ice thickness in the Arctic (Stroeve et al., 2014; 8 Shu et al., 2015) and the observed evolution of albedo on seasonal scales (Koenigk et al., 2014). It was also 9 found that many models’ deviation from observed sea ice cover climatology cannot be explained by internal 10 variability, whereas the models’ deviation from observed sea ice cover trend (over the satellite period) can 11 often be explained by internal variability (Olonscheck and Notz, 2017). This hinders a selection of models 12 according to their simulated trends, which additionally has been shown to only have a weak effect on the 13 magnitude of simulated future trends (Stroeve and Notz, 2015). 14 15 Based on results from the CMIP6 models, we conclude that on average the Arctic will become practically 16 ice-free in September by the end of the 21st century under SSP2-4.5, SSP3-7.0, and SSP5-8.5 (see Figure 17 4.2c and Table 4.4; high confidence). Also, in the CMIP6 models, Arctic SIA in March decreases in the 18 future, but to a much lesser degree, in percentage terms, than in September (Table 4.4; high confidence). A 19 more detailed assessment of projected Arctic and also Antarctic sea ice change can be obtained in Chapter 9, 20 Section 9.3.1. 21 22 23 [START TABLE 4.4 HERE] 24 25 Table 4.4: CMIP6 Arctic sea-ice area for selected months, time periods, and across five SSPs. Displayed are the 26 multi-model averages across the individual models and, in parentheses, the 5–95% ranges. The number of 27 models used in these calculations are shown in Figure 4.2c in Section 4.3.1. 28 Units = 106 km2 SSP1-1.9 SSP1-2.6 SSP2-4.5 SSP3-7.0 SSP5-8.5 September: 2021–2040 2.6 (1.1, 6.5) 2.7 (0.6, 6.4) 2.8 (0.7, 6.4) 3.1 (1.1, 6.4) 2.5 (0.4, 5.8) 2041–2060 2.2 (0.3, 6.5) 2.0 (0.2, 6.1) 1.7 (0.1, 5.6) 1.7 (0.1, 5.7) 1.2 (0.0, 5.2) 2081–2100 2.4 (0.2, 6.2) 1.7 (0.0, 6.0) 0.8 (0.0, 4.6) 0.5 (0.0, 3.3) 0.3 (0.0, 2.2) March: 2021–2040 14.0 (11.4, 18.7) 14.9 (11.9, 25.8) 14.9 (11.9, 23.5) 15.0 (11.7, 27.3) 14.9 (11.9, 24.7) 2041–2060 13.8 (10.9, 18.3) 14.5 (10.9, 25.7) 14.3 (11.1, 23.3) 14.2 (10.5, 27.1) 13.9 (10.2, 24.5) 2081–2100 13.7 (10.9, 18.5) 14.2 (10.6, 25.7) 13.1 (9.5, 22.2) 11.8 (5.4, 25.5) 9.7 (3.1, 21.6) 29 30 [END TABLE 4.4 HERE] 31 32 33 Studies focusing on the relationship of sea-ice extent and changes in external drivers have consistently found 34 a much-reduced likelihood of a practically ice-free Arctic Ocean during summer for global warming of 1.5°C 35 than for 2.0°C (Screen and Williamson, 2017; Jahn, 2018; Niederdrenk and Notz, 2018; Notz and Stroeve, 36 2018; Sigmond et al., 2018; Olson et al., 2019). This is shown here in a large initial-condition ensemble of 37 observationally constrained model simulations where GSAT are stabilized at 1.5°C, 2.0°C and 3.0°C 38 warming relative to 1850–1900 in the RCP8.5 scenario (Figure 4.5). Temperature stabilization is achieved 39 by switching off all the anthropogenic emissions around the time that GSAT first reaches the stabilization 40 thresholds. Simulations have been observationally constrained to correct for a model bias in simulated 41 historical September sea-ice extent. In these simulations, Arctic sea ice coverage in September is simulated, 42 on average, to drop below 1 million km2 around 2040, consistent with the AR5 set of assessed models 43 (Sigmond et al., 2018). The individual model simulations, for which there are twenty for each stabilized 44 temperature level, show that the probability of the Arctic becoming practically ice free at the end of the 21st 45 century is significantly higher for 2°C warming than for 1.5°C warming above 1850–1900 levels (high 46 confidence). Do Not Cite, Quote or Distribute 4-29 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 3 [START FIGURE 4.5 HERE] 4 5 Figure 4.5: Arctic sea-ice extent in September in a large initial-condition ensemble of observationally- 6 constrained simulations of an Earth system model (CanESM2). The black and red curves are average 7 over twenty simulations following historical forcings to 2015 and RCP8.5 extensions to 2100. The 8 coloured curves are averages over twenty simulations each after GSAT has been stabilized at the 9 indicated degree of global mean warming relative to 1850–1900. The bars to the right are the minimum to 10 maximum ranges over 2081–2100 (Sigmond et al., 2018). The horizontal dashed line indicates a 11 practically ice-free Arctic. Further details on data sources and processing are available in the chapter data 12 table (Table 4.SM.1). 13 14 [END FIGURE 4.5 HERE] 15 16 17 4.3.2.2 Global Mean Sea Level 18 19 AR5 assessed from CMIP5 process-based simulations that the rate of GMSL rise during the 21st century will 20 very likely exceed the rate observed during 1971–2010 for all RCP scenarios due to increases in ocean 21 warming and loss of mass from glaciers and ice sheets (Church et al., 2013). Further, AR5 concluded that for 22 the period 2081–2100, compared to 1986–2005, GMSL rise is likely (medium confidence) to be in the 5–95% 23 range of projections from process-based models, which give 0.26–0.55 m for RCP2.6, 0.32–0.63 m for 24 RCP4.5, 0.33–0.63 m for RCP6.0, and 0.45–0.82 m for RCP8.5. For RCP8.5, the rise by 2100 is 0.52–0.98 25 m with a rate during 2081–2100 of 8–16 mm yr-1. 26 27 There have been substantial modelling advances since AR5, with most sea-level projections corresponding to 28 one of three categories: 1) central-range projections, combining scenario-conditional probability distributions 29 for the different contributions to estimate a central range under different scenarios; 2) probabilistic 30 projections, which explicitly consider outcomes for a wide range of likelihoods, including low-likelihood 31 high-impact outcomes, and 3) semi-empirical projections, based on statistical relationships between past 32 GMSL changes and climate variables, which now calibrate individual contributions and are consistent with 33 physical-model based estimates (Chapter 9, Section 9.6.3). 34 35 Based on the assessment of the latest modelling information (see Figure 4.2d and Chapter 9, Section 9.6.3), 36 we conclude that under the SSP3-7.0, the likely range of GMSL change averaged over 2081–2100 relative to 37 1995–2014 is 0.46–0.74 m. Under SSP1-2.6, the likely range over the long-term is 0.30–0.54 m. Further, in 38 SSP2-4.5, SSP3-7.0, and SSP5-8.5, the rise in GMSL is projected to accelerate over the 21st century. A 39 detailed assessment of the processes contributing to these projected rises and accelerations in GMSL, 40 together with a comparison to AR5 and SROCC, can be found in Chapter 9, Section 9.6.3. Projected changes 41 in the thermosteric component of GMSL beyond 2300 are assessed in Section 4.7.1. 42 43 In summary, it is virtually certain that under any one of the assessed SSPs, there will be continued rise in 44 GMSL through the 21st century. 45 46 47 4.3.2.3 Atlantic Meridional Overturning Circulation 48 49 AR5 assessed from CMIP5 simulations that the AMOC will very likely weaken over the 21st century, and 50 the projected weakening of the AMOC is consistent with CMIP5 projections of an increase of high-latitude 51 temperature and high-latitude precipitation, with both effects causing the surface waters at high latitudes to 52 become less dense and therefore more stable (Collins et al., 2013). 53 54 Based on CMIP6 models, we find that over the 21st century, AMOC strength, relative to 1995–2014, shows 55 a multi-model mean decrease in each of the SSP scenarios but with a large spread across the individual 56 simulations (Figure 4.6). We also note that the magnitude of the ensemble-mean strength decrease is Do Not Cite, Quote or Distribute 4-30 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 approximately scenario independent up to about 2060 (Weijer et al., 2020). A more detailed assessment of 2 these projected AMOC changes, and the mechanisms involved, can be found in Chapter 9, Section 9.2.3. 3 4 In summary, we assess from the CMIP6 models that AMOC weakening over the 21st century is very likely; 5 the rate of weakening is approximately independent of the emission scenario (high confidence). 6 7 8 [START FIGURE 4.6 HERE] 9 10 Figure 4.6: CMIP6 annual mean AMOC strength change in historical and scenario simulations. Changes are 11 relative to averages from 1995–2014. The curves show ensemble averages and the shadings the 5–95% 12 ranges across the SSP1-2.6 and SSP3-7.0 ensembles. The circles to the right of the panel show the 13 anomalies averaged from 2081–2100 for each of the available model simulations. The numbers inside the 14 panel are the number of model simulations. Here, the strength of the AMOC is computed as the 15 maximum value of annual-mean ocean meridional overturning mass streamfunction in the Atlantic at 16 26°N. Results are from concentration-driven simulations. Further details on data sources and processing 17 are available in the chapter data table (Table 4.SM.1). 18 19 [END FIGURE 4.6 HERE] 20 21 22 Based on a large initial condition ensemble of simulations with a CMIP5 model (CanESM2) with emission 23 scenarios leading to stabilization of global warming of 1.5°C, 2.0°C, or 3.0°C relative to 1850–1900, AMOC 24 continues to decline for 5–10 years after GSAT is effectively stabilized at the given GWL (Sigmond et al., 25 2020). This is followed by a recovery of AMOC strength for about the next 150 years to a level that is 26 approximately independent of the considered stabilization scenario. These results are replicated in 27 simulations in a CMIP6 model (CanESM5) with emissions cessation after diagnosed CO2 emissions reach 28 750 Gt, 1000 Gt, or 1500 Gt. These emissions levels lead to global warming stabilization at 1.5°C, 2.0°C, or 29 3.0°C relative to 1850–1900. In summary, in these model simulations the AMOC recovers over several 30 centuries after the cessation of CO2 emissions (medium confidence). 31 32 33 4.3.2.4 Ocean and Land Carbon Uptake 34 35 AR5 concluded with very high confidence that ocean carbon uptake of anthropogenic CO2 will continue 36 under all RCPs through the 21century, with higher uptake corresponding to higher concentration pathways. 37 The future evolution of the land carbon uptake was assessed to be much more uncertain than for ocean 38 carbon uptake, with a majority of CMIP5 models projecting a continued cumulative carbon uptake. 39 40 Based on results from the CMIP6 models, we conclude that the flux of carbon from the atmosphere into the 41 ocean increases continually through most of 21st century in the two highest emissions and decreases 42 continually under the other emission scenarios (Figure 4.7a). The flux of carbon from the atmosphere to 43 land shows a similar 21st century behaviour across the scenarios but with much higher year-to-year variation 44 than ocean carbon flux (Figure 4.7b). A more in-depth assessment and discussion of the mechanism involved 45 can be found in Chapter 5, Section 5.4.5. 46 47 In summary, we assess that the cumulative uptake of carbon by the ocean and by land will increase through 48 the 21st century irrespective of the considered emission scenarios (very high confidence). 49 50 51 [START FIGURE 4.7 HERE] 52 53 Figure 4.7: CMIP6 carbon uptake in historical and scenario simulations. (a) Atmosphere to ocean carbon flux 54 (PgC yr-1). (b) Atmosphere to land carbon flux (PgC yr-1). The curves show ensemble averages and the 55 shadings show the 5–95% ranges across the SSP1-2.6 and SSP3-7.0 ensembles. The numbers inside each 56 panel are the number of model simulations. The land uptake is taken as Net Biome Productivity (NBP) Do Not Cite, Quote or Distribute 4-31 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 and so includes any modelled net land-use change emissions. Results are from concentration-driven 2 simulations. Further details on data sources and processing are available in the chapter data table (Table 3 4.SM.1). 4 5 [END FIGURE 4.7 HERE] 6 7 8 4.3.2.5 Surface Ocean pH 9 10 AR5 assessed from CMIP5 simulations that it is virtually certain that increasing storage of carbon by the 11 ocean under all four RCPs through to 2100 will increase ocean acidification in the future (Ciais et al., 2013). 12 Specifically, AR5 reported that CMIP5 models project increased ocean acidification globally to 2100 under 13 all RCPs, and that the corresponding model mean and model spread in the decrease in surface ocean pH from 14 1986–2005 to 2081–2100 would be 0.065 (0.06–0.07) for RCP2.6, 0.145 (0.14–0.15) for RCP4.5, 0.203 15 (0.20–0.21) for RCP6.0 and 0.31 (0.30–0.32) for RCP8.5. 16 17 Based on results from the CMIP6 models we conclude that, except for the lower-emission scenarios SSP1- 18 1.9 and SSP1-2.6, ocean surface pH decreases monotonically through the 21st century (Figure 4.8; high 19 confidence). 20 21 22 [START FIGURE 4.8 HERE] 23 24 Figure 4.8: Global average surface ocean pH. The shadings around the SSP1-2.6 and SSP5-7.0 curves are the 5– 25 95% ranges across those ensembles. The numbers inside each panel are the number of model simulations. 26 Results are from concentration-driven simulations. Further details on data sources and processing are 27 available in the chapter data table (Table 4.SM.1). 28 29 [END FIGURE 4.8 HERE] 30 31 32 4.3.3 Modes of Variability 33 34 4.3.3.1 Northern and Southern Annular Modes 35 36 Northern Annular Mode 37 The Northern Annular Mode (NAM) is the leading mode of variability in the NH extratropical atmosphere 38 (see Annex IV, Section AIV.2.1). Throughout this chapter, we use a simple fixed latitude-based NAM index 39 defined as the difference in SLP between 35°N and 65°N (Li and Wang, 2003; Section AIV.2.1). The NAM 40 index computed from the latitudinal gradient in SLP is strongly correlated with variations in the latitudinal 41 position and strength of the mid-latitude westerly jets, and with the spatial distribution of Arctic sea ice 42 (Caian et al., 2018). Projected changes in the position and strength of the mid-latitude westerly jets, storm 43 tracks, and atmospheric blocking in both hemispheres are assessed in Section 4.5.1.6. AR5 referred to the 44 NAM, and its synonym the Arctic Oscillation (AO), through its regional counterpart, the North Atlantic 45 Oscillation (NAO). Here, we use the term NAM to refer also to the AO and NAO (see Section AIV.2.1), 46 accepting that the AO and NAO are not identical entities. 47 48 We first summarise the assessment of past NAM changes and their attribution from Chapters 2 and 3 to put 49 into context the future projections described here. Strong positive trends for the NAM/NAO indices were 50 observed since 1960, which have weakened since the 1990s (high confidence; Chapter 2, Section 2.4.1.1). 51 The NAO variability in the instrumental record was likely not unusual in the millennial and multi-centennial 52 context (Section 2.4.1.1). Climate models simulate the gross features of the NAM with reasonable fidelity, 53 including its interannual variability, but models tend to systematically underestimate the amount of 54 multidecadal variability of the NAM and jet stream compared to observations (Wang et al., 2017b; 55 Bracegirdle et al., 2018; Simpson et al., 2018a); Chapter 3, Section 3.7.1), with the caveat of the 56 observational record being relatively short to characterise decadal variability (Chiodo et al., 2019). A Do Not Cite, Quote or Distribute 4-32 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 realistic simulation of the stratosphere and SST variability in the tropics and northern extratropics are 2 important for a model to realistically capture the observed NAM variability. Despite some evidence from 3 climate model studies that anthropogenic forcings influence the NAM, the balance of evidence indicates that 4 there is little evidence for a significant role for anthropogenic forcings in driving the observed multidecadal 5 variations of the NAM over the instrumental period (Section 3.7.1). 6 7 AR5 assessed from CMIP5 simulations that the future boreal wintertime NAM is very likely to exhibit large 8 natural variations and trends of similar magnitude to that observed in the past and is likely to become slightly 9 more positive in the future (Collins et al., 2013). Based on CMIP6 model results displayed in Figure 4.9a, we 10 conclude that the boreal wintertime surface NAM is more positive by the end of the 21st century under 11 SSP3-7.0 and SSP5-8.5 (high confidence). For these high emission scenarios, the 5–95% range of NAM 12 index anomalies averaged from 2081–2100 are 0.3–3.8 hPa and 0.32–5.2 hPa, respectively. On the other 13 hand, under neither of the lowest emission scenarios, SSP1-1.9 and SSP1-2.6, does the NAM show a robust 14 change, by the end of the 21st century (high confidence). 15 16 17 [START FIGURE 4.9 HERE] 18 19 Figure 4.9: CMIP6 simulations of boreal wintertime (DJF) Annular Mode indices. (a) NAM and (b) SAM. The 20 NAM is defined as the difference in zonal mean SLP at 35°N and 65°N (Li and Wang, 2003) and the 21 SAM as the difference in zonal mean SLP at 40°S and 65°S (Gong and Wang, 1999). All anomalies are 22 relative to averages from 1995–2014. The curves show multi-model ensemble averages over the CMIP6 23 r1 simulations. The shadings around the SSP1-2.6 and SSP3-7.0 curves denote the 5–95% ranges of the 24 ensembles. The numbers inside each panel are the number of model simulations. The results are for 25 concentration-driven simulations. Further details on data sources and processing are available in the 26 chapter data table (Table 4.SM.1). 27 28 [END FIGURE 4.9 HERE] 29 30 31 Significant progress has been made since AR5 in understanding the physical mechanisms responsible for 32 changes in the NAM, although uncertainties remain. It is now clear from the literature that the NAM 33 response, and the closely-related response of the mid-latitude storm tracks, to anthropogenic forcing in 34 CMIP5-era climate models is determined by a ‘tug-of-war’ between two opposing processes (Harvey et al., 35 2014; Shaw et al., 2016a; Screen et al., 2018a); 1) Arctic amplification (see Section 4.5.1.1 and Chapter 7, 36 Section 7.4.4.1), which decreases the low-level meridional temperature gradient, reduces baroclinicity on the 37 poleward flank of the eddy-driven jet, and shifts the storm tracks equatorward and leading to a negative 38 NAM (see Box 10.1; Harvey et al., 2015; Hoskins and Woollings, 2015; Peings et al., 2017; Screen et al., 39 2018a); 2) and enhanced warming in the tropical upper-troposphere, due to GHG increases and associated 40 water vapour and lapse rate feedbacks, which increases the upper-level meridional temperature gradient and 41 causes a poleward shift of the storm tracks and a positive NAM (Harvey et al., 2014; Vallis et al., 2015; 42 Shaw, 2019). The large diversity in projected NAM changes in CMIP5 multi-model ensemble (Gillett and 43 Fyfe, 2013) appears to be at least partly explained by the relative importance of these two mechanisms in 44 particular models (Harvey et al., 2014, 2015; Vallis et al., 2015; McCusker et al., 2017; Oudar et al., 2017). 45 Models that produce larger Arctic amplification also tend to produce larger equatorward shifts of the mid- 46 latitude jets and associated negative NAM responses (Barnes and Polvani, 2015; Harvey et al., 2015; Zappa 47 and Shepherd, 2017; McKenna et al., 2018; Screen et al., 2018a; Zappa et al., 2018). 48 49 Another area of progress is new understanding the role of cloud radiative effects in shaping the mid-latitude 50 circulation response to anthropogenic forcing. Through their non-uniform distribution of radiative heating, 51 cloud changes can modify meridional temperature gradients and alter mid-latitude circulation and the 52 annular modes in both hemispheres (Ceppi et al., 2014; Voigt and Shaw, 2015, 2016; Ceppi and Hartmann, 53 2016; Ceppi and Shepherd, 2017; Lipat et al., 2018; Albern et al., 2019; Voigt et al., 2019). In addition to the 54 effects of changing upper and lower tropospheric temperature gradients on the NAM, progress has been 55 made since AR5 in understanding the effect of simulated changes in the strength of the stratospheric polar 56 vortex on winter NAM projections (Manzini et al., 2014; Zappa and Shepherd, 2017; Simpson et al., 2018b). Do Not Cite, Quote or Distribute 4-33 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 Southern Annular Mode 3 The Southern Annular Mode (SAM) is the leading mode of large-scale extratropical atmospheric variability 4 in the SH and influences most of the southern extratropics (see Annex IV, Section AIV.2.2). In its positive 5 phase, the SAM characterizes anomalously low pressure over the polar cap and high pressure in southern 6 mid-latitudes (Marshall, 2003). While there are some zonal asymmetries to the structure of the SAM (see 7 Section AIV.2.2), it is more symmetric than its NH counterpart (Fyfe et al., 1999). Throughout this chapter, 8 we use a simple fixed latitude-based SAM index defined as the difference in zonal mean SLP between 40°S 9 and 65°S (Gong and Wang, 1999) (see Section AIV.2.2 for discussion of other SAM indices). Although the 10 SAM is often used as a proxy for the location of the mid-latitude westerly wind belt, trends in the SAM can 11 reflect a combination of changes in jet position, width, and strength. The changes in the SH circulation 12 associated with the SAM influence on surface wind stress (Wang et al., 2014) and hence affect the Southern 13 Ocean. 14 15 Over the instrumental period, there has been a robust positive trend in the SAM index, particularly since 16 1970 (high confidence; Chapter 2, Section 2.4.1.2). There is medium confidence that the recent trend in the 17 SAM is unprecedented in the past several Centuries (Section 2.4.1.2). There is high confidence that 18 stratospheric ozone depletion and GHG increases have contributed to the positive SAM trend during the late 19 20th century, with ozone depletion dominating in austral summer, following the peak of the Antarctic ozone 20 hole in September–October, and GHG increases dominating in other seasons (Chapter 3, Section 3.7.2). To 21 capture the effects of stratospheric ozone changes on the SAM, climate models must include a realistic 22 representation of ozone variations (Section 3.7.2). In models that do not explicitly represent stratospheric 23 ozone chemistry, which includes the majority of the CMIP6 model ensemble, an ozone dataset is prescribed. 24 To properly capture the effects of ozone depletion and recovery on the stratosphere and surface climate, the 25 prescribed ozone dataset must realistically capture observed stratospheric ozone trends with sufficiently high 26 temporal resolution (Neely et al., 2014; Young et al., 2014). The CMIP6 experiment protocol recommended 27 the use of a prescribed 4-D monthly mean ozone concentration field for models without stratospheric 28 chemistry (Eyring et al., 2016). 29 30 AR5 assessed that the positive trend in the austral summer/autumn SAM observed since 1970 (see Section 31 2.4.1.2) is likely to weaken considerably as stratospheric ozone recovers through the mid-21st century, while 32 in other seasons the SAM changes depend on the emission scenario, with a larger increase in SAM for higher 33 emission scenarios. In CMIP6 models, the austral summer SAM is more positive by the end of the 21st 34 century under SSP3-7.0 and SSP5-8.5 (Figure 4.9b). On the other hand, under SSP1-1.9 and SSP1-2.6, the 35 SAM is projected to be less positive, especially under SSP1-1.9 where the 5–95% ranges of anomalies 36 relative to 1995–2014 are –3.1 to 0.0 hPa averaged from 2081–2100. In summary, under the highest 37 emission scenarios in the CMIP6 models, the SAM in the austral summer becomes more positive through the 38 21st century (high confidence). 39 40 41 4.3.3.2 El Niño-Southern Oscillation 42 43 ENSO is the most dominant mode of variability on interannual timescales and also the dominant source of 44 seasonal climate predictability (Timmermann et al., 2018; see Chapter 11, Box 11.3 and Annex IV, Section 45 AIV.2.3). AR5 assessed from CMIP5 simulations that ENSO variability will very likely remain the dominant 46 mode of interannual climate variability in the future, and that associated ENSO precipitation variability on 47 regional scales is likely to intensify (Christensen et al., 2013). However, they assessed there was low 48 confidence in projected changes in ENSO variability in the 21st century due to a strong component of 49 internal variability. 50 51 52 [START FIGURE 4.10 HERE] 53 54 Figure 4.10: Changes in amplitude of ENSO Variability. Variability of (a) SST and (b) precipitation anomalies 55 averaged over Niño3.4 region for 1950–2014 from CMIP6 historical simulations and for 2015–2100 from Do Not Cite, Quote or Distribute 4-34 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 four SSPs. Thick lines stand for multi-model mean and shading is the 5–95% range across CMIP6 models 2 for historical simulation (grey), SSP1-2.6 (blue) and SSP3-7.0 (pink), respectively. The amplitude of 3 ENSO SST and rainfall variability is defined as the standard deviation of the detrended Niño3.4-area 4 averaged SST and rainfall index, respectively, over 30-year running windows. The standard deviation in 5 every single model is normalized by each model’s present-day standard deviation averaged from 1995 to 6 2014. The number of available models is listed in parentheses. This figure is adopted from (Yun et al., 7 2021). Further details on data sources and processing are available in the chapter data table (Table 8 4.SM.1). 9 10 [END FIGURE 4.10 HERE] 11 12 13 Among a range of indices proposed for representing ENSO, we use the most prominent one, the Niño 3.4 14 index, defined as the average equatorial SST or precipitation across the central equatorial Pacific (5°S–5°N, 15 170°W–120°W) (see Section AIV.2.3). Here, we consider the evolution of the amplitude of Niño3.4 index 16 for SST and precipitation over the 21st century as projected by CMIP6 models. Analysis of CMIP6 models 17 shows there is no robust model consensus on the forced changes in the amplitude of ENSO SST variability 18 even under the high-emission scenarios SSP3-7.0 and SSP5-8.5, but a significant increasing trend in the 19 amplitude of ENSO precipitation variability is projected across the 21st century in the four SSPs (Figure 20 4.10). This is broadly consistent with results from CMIP5 models (Christensen et al., 2013)(Power et al., 21 2013)(Cai et al., 2015)(Chen et al., 2017)(Wengel et al., 2018), recent studies with CMIP6 models (Brown et 22 al., 2020)(Fredriksen et al., 2020a)(Freund et al., 2020a)(Yun et al., 2021), and large initial-condition 23 ensemble experiments (Maher et al., 2018; Zheng et al., 2018; Haszpra et al., 2020). 24 25 It is therefore very likely that the amplitude of ENSO rainfall variability will intensify in response to global 26 warming over the 21st century although there is no robust consensus from CMIP6 climate models for a 27 systematic change in amplitude of ENSO SST variability even in the high-emission scenarios of SSP3-7.0 28 and SSP5-8.5. 29 30 31 4.3.4 Synthesis Assessment of Projected Change in Global Surface Air Temperature 32 33 GSAT change is assessed using multiple lines of evidence including the CMIP6 projection simulations out to 34 year 2100. The assessment combines CMIP6 projections driven by SSP scenarios with observational 35 constraints on simulated past warming (Brunner et al., 2020; Liang et al., 2020; Nijsse et al., 2020; Tokarska 36 et al., 2020; Ribes et al., 2021) (see BOX 4.1 and Figure 4.11:a,b), as well as the AR6-updated assessment of 37 ECS and TCR in Section 7.5. The approaches of (Liang et al., 2020; Tokarska et al., 2020; Ribes et al., 2021) 38 have first been extended to all 20-year averaging periods between 2000 and 2100. For each 20-year period, 39 the 5%, 50%, and 95% percentile GSAT values of these three constrained CMIP6 results are averaged 40 percentile by percentile (Figure 4.11c). Then, an emulator based on a two-layer energy balance model (e.g., 41 Held et al., 2010) is driven by the Chapter 7-derived ERF. The emulator parameters are chosen such that the 42 central estimate, lower bound of the very likely range, and upper bound of the very likely range of climate 43 feedback parameter and ocean heat uptake coefficient take the values that map onto the corresponding 44 combination of ECS (3°C, 2°C, and 5°C, respectively) and TCR (1.8°C, 1.2°C, and 2.4°C, respectively) of 45 Section 7.5 (see Box 4.1). As a final step, the constrained-CMIP6 and the emulator-based 5%, 50%, and 95% 46 percentile GSAT values are averaged percentile by percentile (Figure 4.11c, d; Table 4.5). Constrained 47 CMIP6 results and the ECS- and TCR-based emulator thus contribute one-half each to the GSAT 48 assessment. Because the emulator results and (Ribes et al., 2021) represent the forced response only, and 49 averaging over the other two individual estimates (Liang et al., 2020; Tokarska et al., 2020) further reduces 50 the contribution from internal variability, the assessed GSAT time series are assumed to represent purely the 51 forced response. 52 53 Averaged over the period 2081–2100, GSAT is very likely to be higher than in the recent past (1995–2014) 54 by 0.3°C–0.9°C in the low-emission scenario SSP1-1.9 and by 2.6°C–4.7°C in the high-emission scenario 55 SSP5-8.5. For the scenarios SSP1-2.6, SSP2-4.5, and SSP3-7.0, the corresponding very likely ranges are 56 0.6°C–1.4°C, 1.3°C–2.5°C, and 2.0°C–3.8°C, respectively (Figure 4.11, Table 4.5). Because the different Do Not Cite, Quote or Distribute 4-35 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 approaches for estimating long-term GSAT change produce consistent results (Figure 4.11), there is high 2 confidence in this assessment. These ranges of the long-term projected GSAT change generally correspond 3 to AR5 ranges for related scenarios but the likelihood is increased to very likely ranges, in contrast to the 4 likely ranges in AR5. Over the mid-term period 2041–2060, the very likely GSAT ranges of SSP1-1.9 and 5 SSP5-8.5 are almost completely distinct (Table 4.5, high confidence, see also Section 4.3.1). 6 7 CMIP6 models project a wider range of GSAT change than the assessed range (high confidence, see Section 8 4.3.1). The CMIP6 models with a higher climate sensitivity simulate warming rates higher than assessed very 9 likely here (see Section 4.3.1); these rates are very unlikely but not impossible to occur and hence cannot be 10 excluded. The implications of these very unlikely warming rates for patterns of surface temperature and 11 precipitation change are assessed in Section 4.8. 12 13 For the near term, initialized decadal forecasts constitute another line of evidence over the period 2019–2028 14 (see Box 4.1). The forecasts are consistent with the assessed GSAT very likely range (Box 4.1, Figure 1), 15 strengthening the confidence in the near-term assessment. 16 17 18 [START FIGURE 4.11 HERE] 19 20 Figure 4.11: Multiple lines of evidence for GSAT changes for the long-term period, 2081–2100, relative to the 21 average over 1995–2014, for all five priority scenarios. The unconstrained CMIP6 5–95% ranges 22 (coloured bars) in (a) differ slightly because different authors used different subsamples of the CMIP6 23 archive. The constrained CMIP6 5–95% ranges (coloured bars) in (b) are smaller than the unconstrained 24 ranges in (a) and differ because of different samples from the CMIP6 archive and because different 25 observations and methods are used. In (c), the average of the ranges in (b) is formed (grey bars). Green 26 bars in (c) show the emulator ranges, defined such that the best estimate, lower bound of the very likely 27 range, and upper bound of the very likely range of climate feedback parameter and ocean heat uptake 28 coefficient take the values that map onto the corresponding values of ECS and TCR of Section 7.5 (see 29 BOX 4.1). The time series in (d) are constructed by taking the average of the constrained CMIP6 ranges 30 and the emulator ranges. The y-axes on the right-hand side are shifted upward by 0.85°C, the central 31 estimate of the observed warming for 1995–2014, relative to 1850–1900 (Cross-Chapter Box 2.3, Table 32 1). Further details on data sources and processing are available in the chapter data table (Table 4.SM.1). 33 34 [END FIGURE 4.11 HERE] 35 36 37 The assessed ranges of GSAT change can be converted to change relative to mean GSAT over the period 38 1850–1900 for a consistent comparison with AR5 (IPCC, 2013) and SR1.5 (Masson-Delmotte et al., 2018). 39 GSAT was warmer in 1995–2014 (recent past) than 1850–1900 by 0.85°C [0.67–0.98°C]. GSAT diagnosed 40 for 1986–2005 (AR5 recent past) relative to 1850–1900 is 0.08°C higher than was diagnosed in AR5, due to 41 methodological and dataset updates (Cross-Chapter Box 2.3, Table 1). 42 43 The uncertainty in GSAT relative to 1850–1900 includes the very likely ranges of assessed GSAT change 44 relative to 1995–2014 (depending on scenario and period, between 0.5°C and 2.4°C, Figure 4.11d, Table 45 4.5), the uncertainty in historical GSAT change from the mean over 1850–1900 to 1995–2014 (about 0.3°C, 46 Cross-Chapter Box 2.3), and the estimate of internal variability in 20-year GSAT averages (5–95% range 47 about 0.15°C, Box 4.1, (Maher et al., 2019a)). These uncertainties are assumed to be independent and are 48 added in quadrature, meaning that the total uncertainty is only slightly larger than the dominating 49 contribution by the GSAT change relative to 1995–2014 (Table 4.5). The addition is done by numerically 50 sampling a normal distribution fitted to the 5%, 50%, and 95% percentiles of the internal variability, as well 51 as sampling skew-normal distributions (e.g., O’Hagan and Leonard, 1976) fitted to the 5%, 50%, and 95% 52 percentiles of both historical warming and GSAT relative to 1995–2014. The result is a joint probability 53 distribution of GSAT change and 20-year period. 54 55 Averaged over the period 2081–2100, GSAT is very likely to be higher than in the period 1850–1900 by 56 1.0°C–1.8°C in the low-emission scenario SSP1-1.9 and by 3.3°C–5.7°C in the high-emission scenario Do Not Cite, Quote or Distribute 4-36 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 SSP5-8.5. For the scenarios SSP1-2.6, SSP2-4.5, and SSP3-7.0, the corresponding very likely ranges are 2 1.3°C–2.4°C, 2.1°C–3.5°C, and 2.8°C–4.6°C, respectively (Table 4.5). 3 4 Time series of assessed GSAT change are now used to assess the time when certain thresholds of GSAT 5 increase are crossed (Table 4.5). The threshold-crossing time is defined as the midpoint of the first 20-year 6 period during which the average GSAT exceeds the threshold. During the near term (2021–2040), a 1.5°C 7 increase in the 20-year average of GSAT, relative to the average over the period 1850–1900, is very likely to 8 occur in scenario SSP5-8.5, likely to occur in scenarios SSP2-4.5 and SSP3-7.0, and more likely than not to 9 occur in scenarios SSP1-1.9 and SSP1-2.6. In all scenarios assessed here except SSP5-8.5, the central 10 estimate of crossing the 1.5°C threshold lies in the early 2030s, about ten years earlier than the midpoint of 11 the likely range (2030–2052) assessed in the SR1.5, which assumed continuation of the then-current warming 12 rate. Roughly half of this ten-year difference arises from a larger historical warming diagnosed in AR6, 13 while the other half arises because for central estimates of climate sensitivity, most scenarios show stronger 14 warming over the near term than was estimated as ‘current’ in SR1.5 (medium confidence); this estimate has 15 been confirmed in AR6 (Section 3.3.1). If ECS and TCR lie near the lower end of the assessed very likely 16 range, crossing the 1.5°C warming threshold is avoided in scenarios SSP1-1.9 and SSP1-2.6 (medium 17 confidence). It is more likely than not that under SSP1-1.9, GSAT relative to 1850–1900 will remain below 18 1.6°C throughout the 21st century, implying a potential temporary overshoot above 1.5°C of no more than 19 0.1°C. All statements about crossing the 1.5°C threshold assume that no major volcanic eruption occurs 20 during the near term. 21 22 A warming level of 2°C in GSAT, relative to the period 1850–1900, is very likely to be crossed in the mid- 23 term period 2041–2060 under SSP5-8.5, likely to be crossed in the mid-term period under SSP3-7.0, and 24 more likely than not to be crossed during the mid-term period under SSP2-4.5. During the entire 21st 25 century, a warming level of 2°C in GSAT, relative to the period 1850–1900, will be crossed under SSP5-8.5 26 and SSP3-7.0, will extremely likely be crossed under SSP2-4.5, will unlikely be crossed under SSP1-2.6, and 27 will extremely unlikely be crossed under SSP1-1.9. 28 Do Not Cite, Quote or Distribute 4-37 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 [START TABLE 4.5 HERE] 3 4 Table 4.5: Assessment results for 20-year averaged GSAT change, based on multiple lines of evidence. The change is displayed in °C relative to the 1995–2014 5 and 1850–1900 reference periods for selected time periods (near term 2021–2040, mid-term 2041–2060, and long term 2081–2100), and as the time when 6 certain temperature thresholds are crossed, relative to the period 1850–1900. The recent reference period 1995–2014 was higher in GSAT than the period 7 1850–1900 by 0.85°C [0.67–0.98°C], (Cross-Chapter Box 2.3). The entries give both the central estimate and, in parentheses, the very likely (5–95%) 8 range. An entry n.c. means that the global warming threshold is not crossed during the period 2021–2100. 9 SSP1-1.9 SSP1-2.6 SSP2-4.5 SSP3-7.0 SSP5-8.5 Near term, 2021–2040 relative to 1995–2014 0.6 (0.4, 0.9) 0.6 (0.4, 0.9) 0.7 (0.4, 0.9) 0.7 (0.4, 0.9) 0.8 (0.5, 1.0) relative to 1850–1900 1.5 (1.2, 1.7) 1.5 (1.2, 1.8) 1.5 (1.2, 1.8) 1.5 (1.2, 1.8) 1.6 (1.3, 1.9) Mid-term, 2041–2060 relative to 1995–2014 0.7 (0.4, 1.1) 0.9 (0.5, 1.3) 1.1 (0.8, 1.6) 1.3 (0.9, 1.7) 1.5 (1.1, 2.1) relative to 1850–1900 1.6 (1.2, 2.0) 1.7 (1.3, 2.2) 2.0 (1.6, 2.5) 2.1 (1.7, 2.6) 2.4 (1.9, 3.0) Long term, 2081–2100 relative to 1995–2014 0.6 (0.2, 1.0) 0.9 (0.5, 1.5) 1.8 (1.2, 2.6) 2.8 (2.0, 3.7) 3.5 (2.4, 4.8) relative to 1850–1900 1.4 (1.0, 1.8) 1.8 (1.3, 2.4) 2.7 (2.1, 3.5) 3.6 (2.8, 4.6) 4.4 (3.3, 5.7) 1.5°C, 2025–2044 2023–2042 2021–2040 2021–2040 2018–2037 relative to 1850–1900 (2013–2032, n.c.) (2012–2031, n.c.) (2012–2031, 2037–2056) (2013–2032, 2033–2052) (2011–2030, 2029–2048) 2°C, n.c. n.c. 2043–2062 2037–2056 2032–2051 relative to 1850–1900 (n.c., n.c.) (2031–2050, n.c.) (2028–2047, 2075–2094) (2026–2045, 2053–2072) (2023–2042, 2044–2063) 3°C, n.c. n.c. n.c. 2066–2085 2055–2074 relative to 1850–1900 (n.c., n.c.) (n.c., n.c.) (2061–2080, n.c.) (2050–2069, n.c.) (2042–2061, 2074–2093) 4°C, n.c. n.c. n.c. n.c. 2075–2094 relative to 1850–1900 (n.c., n.c.) (n.c., n.c.) (n.c., n.c.) (2070–2089, n.c.) (2058–2077, n.c.) 10 11 12 [END TABLE 4.5 HERE] Do Not Cite, Quote or Distribute 4-38 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 3 4.4 Near-term Global Climate Changes 4 5 This section assesses changes in large-scale climate over the period 2021–2040 and includes information 6 from both projections and initialized decadal predictions. The structure is similar to Section 4.3. Unless 7 noted otherwise, the assessment assumes that there will be no major volcanic eruption in the near term. The 8 climatic effects of volcanic eruptions are assessed in Section 4.4.4 and Cross-Chapter Box 4.1; Section 4.4.4 9 also assesses the climate effects of short-lived climate forcers. 10 11 12 4.4.1 Atmosphere 13 14 4.4.1.1 Average Global Surface Air Temperature 15 16 AR5 assessed that it is likely that GSAT will increase in the range 0.3°C–0.7°C over the period 2016–2035 17 relative to 1986–2005 (medium confidence), and that there were not large differences in the GSAT change 18 among different RCPs in this period (Kirtman et al., 2013). AR5 further assessed that it is more likely than 19 not that the mean GSAT for the period 2016–2035 will be more than 1°C above the mean for 1850–1900, 20 and it is very unlikely that it will be more than 1.5°C above the 1850–1900 mean (medium confidence). It 21 was shown that in the period 2016–2035, differences in GSAT across RCP scenarios for a single climate 22 model are typically smaller than differences between climate models under a single RCP scenario, indicating 23 that model structural uncertainty is larger than scenario uncertainty over that period (Hawkins and Sutton, 24 2009). 25 26 Near-term (2021–2040) GSAT changes relative to 1995–2014 exhibit only minor dependence on SSP 27 scenario, consistent with AR5 (Table 4.5). Averaged over the twenty years of the near term and across all 28 scenarios, GSAT is very likely to be higher than over 1995–2014 by 0.4°C–1.0°C (Table 4.5), with most of 29 the uncertainty arising from that in ECS and TCR (high confidence, e.g., Lehner et al., 2020, Section 4.3.4). 30 The assessed near-term warming is thus larger than in AR5 by 0.1°C to 0.2°C. This upward revision has the 31 same magnitude as the ad-hoc downward adjustment to near-term projected GSAT change that was 32 performed in AR5 ((Kirtman et al., 2013); see Box 4.1). 33 34 Averaged near-term GSAT is as likely as not at least 1.5°C higher than during 1850–1900, across the five 35 SSP scenarios used here (Table 4.5, see Section 4.3.4). This much higher likelihood of near-term warming 36 reaching 1.5°C than in AR5 arises both because surface warming has continued since AR5 (the period 1995– 37 2014 was warmer by 0.16°C than 1986–2005, Cross-Chapter Box 2.3, Table 1), and because of 38 methodological and dataset updates (the AR6 assessment of 1986–2005 GSAT change relative to 1850–1900 39 is 0.08°C higher than in the AR5; Cross-Chapter Box 2.3, Table 1). 40 41 For annual-mean GSAT, uncertainty in near-term projections arises in roughly equal measure from internal 42 variability and model uncertainty (high confidence, Box 4.1). Forecasts initialized from recent observations 43 simulate GSAT changes for the period 2019–2028 relative to the recent past that are consistent with the 44 assessed very likely range in annual-mean GSAT (Box 4.1, Figure 1, Table 4.5, high confidence). Because 45 annual mean GSAT shows a higher level of internal variability than the 20-year mean, individual years are 46 expected to cross the 1.5°C earlier than the assessed GSAT does. For example, Smith et al. (2018) apply a 47 multi-model decadal-forecast ensemble to assess the likelihood that global warming of 1.5°C higher than 48 over 1850–1900 will be temporarily exceeded in the near future. 49 50 When we repeat the uncertainty quantification for GSAT as in Section 4.3.4 but with the corresponding 51 higher level of internal variability for annual instead of 20-year averages added in quadrature, we can 52 estimate the likelihood that an individual year would cross the GSAT 1.5°C threshold. By 2030, GSAT in 53 any individual year could exceed 1.5°C relative to 1850–1900 with a likelihood between 40 and 60 percent, 54 across the scenarios considered here (medium confidence). 55 Do Not Cite, Quote or Distribute 4-39 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 4.4.1.2 Spatial Patterns of Surface Warming 3 4 Consistent with AR5 and earlier assessments, Figure 4.12 shows for SSP1-2.6 and SSP3-7.0 that the largest 5 warming occurs at high latitudes, particularly in boreal winter in the Arctic (see Section 4.5.1.1), and larger 6 warming over land than over the ocean (see also Section 4.5.1.1). In both scenarios, the increase in seasonal 7 mean surface temperatures over many NH land regions exceeds 1°C relative to 1995–2014. In the near term, 8 the two scenarios show surface temperature changes that are similar in magnitude. The trajectories for well- 9 mixed GHGs, and as a consequence the effective radiative forcing, in the scenarios have not yet diverged 10 that much (O’Neill et al., 2016; Riahi et al., 2017). Based on the currently available CMIP6 models, regions 11 that do not show robust warming in the near-term include the northern North Atlantic, parts of India, parts of 12 North America and Eurasia in winter, and the subtropical eastern Pacific in the Southern Hemisphere. 13 14 15 [START FIGURE 4.12 HERE] 16 17 Figure 4.12: Near-term change of seasonal mean surface temperature. Displayed are projected spatial patterns of 18 CMIP6 multi-model mean change (°C) in (top) DJF and (bottom) JJA near-surface air temperature for 19 2021–2040 from SSP1-2.6 and SSP3-7.0 relative to 1995–2014. The number of models used is indicated 20 in the top right of the maps. No overlay indicates regions where the change is robust and likely emerges 21 from internal variability, that is, where at least 66% of the models show a change greater than the 22 internal-variability threshold (see Section 4.2.6) and at least 80% of the models agree on the sign of 23 change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 24 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate 25 areas of conflicting signals where at least 66% of the models show change greater than the internal- 26 variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data 27 sources and processing are available in the chapter data table (Table 4.SM.1). 28 29 [END FIGURE 4.12 HERE] 30 31 32 The ERF patterns from aerosols and well-mixed GHGs are distinct (Chapter 7), and warming patterns 33 therefore depend on the precise mix of forcing agents in the scenarios. The spatial efficacies – the change in 34 surface temperature per unit ERF – for CO2, sulphate and black carbon aerosols and solar forcing have been 35 recently evaluated in climate models (Modak et al., 2016; Duan et al., 2018; Modak et al., 2018a; Modak and 36 Bala, 2019; Richardson et al., 2019). On average, the spatial patterns of near-surface warming are largely 37 similar for different external drivers (Xie et al., 2013; Richardson et al., 2019; Samset et al., 2020), despite 38 the patterns of forcing being different and large spread across different models (Richardson et al., 2019). 39 40 Internal variability in near-surface temperature change is large in many regions, particularly in mid-latitudes 41 and polar regions (Hawkins and Sutton, 2012). Projections from individual realizations can therefore exhibit 42 divergent regional responses in the near-term in areas where the amplitude of a forced signal is relatively 43 small compared to internal variability (Deser et al., 2012b, 2014, 2016). 44 45 46 4.4.1.3 Precipitation 47 48 AR5 assessed that zonal mean precipitation will very likely increase in high and some of the mid latitudes 49 and will more likely than not decrease in the subtropics. AR5 further assessed that the near-term changes in 50 precipitation are largely uncertain at regional scales, and much of the non-robustness in near-term 51 projections is attributable to internal variability and model uncertainty. 52 53 The mean patterns of seasonal precipitation change in CMIP6 models are consistent with AR5, increasing at 54 high latitudes, over oceanic regions, and in wet regions over the tropics; and decreasing in dry regions 55 including large parts of the subtropics (Figure 4.13). The magnitude of projected changes in precipitation in 56 the near term, especially on regional scales is small compared to the magnitude of internal variability Do Not Cite, Quote or Distribute 4-40 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 (Hawkins and Sutton, 2011, 2016; Hoerling et al., 2011; Deser et al., 2012b; Power et al., 2012) (see Section 2 10.4.3). Analyses of CMIP5, CMIP6, and single-model large-ensemble simulations show that for the 3 uncertainty in near-term precipitation projections, model uncertainty and internal variability dominate while 4 the scenario uncertainty is very small (Lehner et al., 2020)(also see Section 8.5). Based on large ensembles 5 of climate change experiments, it was shown that internal variability decreases over time for both 6 temperature and precipitation on decadal scales (Zhang and Delworth, 2018; Tebaldi et al., 2021). The 7 precipitation projections from CMIP6 models shows larger model uncertainty associated with the higher 8 average transient climate response (Lehner et al., 2020). 9 10 The ‘wet get wetter, dry get drier’ paradigm, which has been used to explain the global precipitation pattern 11 responding to global warming (Held and Soden, 2006a), might not hold, especially over subtropical land 12 regions (Greve et al., 2014; Feng and Zhang, 2015; Greve and Seneviratne, 2015). Over the tropical oceans, 13 precipitation changes are largely driven by the pattern of SST changes (He et al., 2018), and in the 14 subtropics, precipitation response is driven primarily by the fast adjustment to CO2 forcing (He and Soden, 15 2017). In addition to the response to GHG forcing, forcing from natural and anthropogenic aerosols exert 16 impacts on regional patterns of precipitation (Ramanathan et al., 2005; Bollasina et al., 2011; Polson et al., 17 2014; Krishnan et al., 2016; Liu et al., 2018b; Shawki et al., 2018) (also see Section 10.3.1). The large 18 uncertainties in near-term regional precipitation projections arise due to the interplay between internal 19 variability and anthropogenic external forcing (Endo et al., 2018; Wang et al., 2021). Uncertainties in future 20 aerosol emission scenarios contribute to uncertinites in regional precipitation projections (Wilcox et al., 21 2020). Aerosol changes induce a drying in the SH tropical band compensated by wetter conditions in the NH 22 counterpart (Acosta Navarro et al., 2017). The spatially uneven distribution of the aerosol forcing may also 23 induce changes in tropical precipitation caused by shifts in the mean location of the intertropical 24 convergence zone (ITCZ) (Hwang et al., 2013; Ridley et al., 2015; Voigt et al., 2017). Because of the large 25 uncertainty in the aerosol radiative forcing and the dynamical response to the aerosol forcing there is medium 26 confidence in the impacts of aerosols on near-term projected changes in precipitation. Precipitation changes 27 in the near term show seasonal amplification, precipitation increase in the rainy season and decrease in the 28 dry season (Fujita et al., 2019). 29 30 31 [START FIGURE 4.13 HERE] 32 33 Figure 4.13: Near-term change of seasonal mean precipitation. Displayed are projected spatial patterns of CMIP6 34 multi-model mean change (%) in (top) DJF and (bottom) JJA precipitation from SSP1-2.6 and SSP3-7.0 35 in 2021–2040 relative to 1995–2014. The number of models used is indicated in the top right of the maps. 36 No overlay indicates regions where the change is robust and likely emerges from internal variability, that 37 is, where at least 66% of the models show a change greater than the internal-variability threshold (see 38 Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions 39 with no change or no robust significant change, where fewer than 66% of the models show change greater 40 than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 41 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all 42 models agree on the sign of change. Further details on data sources and processing are available in the 43 chapter data table (Table 4.SM.1). 44 45 46 [END FIGURE 4.13 HERE] 47 48 49 Consistent with AR5, we conclude that projected changes of seasonal mean precipitation in the near term 50 will increase at high latitudes. Near-term projected changes in precipitation are uncertain mainly because of 51 natural internal variability, model uncertainty, and uncertainty in natural and anthropogenic aerosol forcing 52 (medium confidence). 53 54 55 4.4.1.4 Global Monsoon Precipitation and Circulation 56 Do Not Cite, Quote or Distribute 4-41 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 The global monsoon is a forced response of the coupled atmosphere-land-ocean system to the annual cycle 2 of solar insolation and characterized by a seasonal reversal of circulation and a seasonal alternation of dry 3 and wet conditions (see Chapter 8, Section 8.3.2, Figure 8.11; Annex V). The global monsoon concept helps 4 to dissect the mechanisms and controlling factors of monsoon variability at various temporal-spatial scales 5 (Wang and Ding, 2008; Wang et al., 2017c). 6 7 8 [START FIGURE 4.14 HERE] 9 10 Figure 4.14: Time series of global land monsoon precipitation and Northern Hemisphere summer monsoon 11 (NHSM) circulation index anomalies. (a) Global land monsoon precipitation index anomalies (Unit: %) 12 defined as the area-weighted mean precipitation rate in the global land monsoon domain defined by Wang 13 et al. (2013) for the CMIP6 historical simulation for 1950–2014 and five SSPs 2015–2100. (b) Anomalies 14 in NHSM circulation index (Unit: m s-1), defined as the vertical shear of zonal winds between 850 and 15 200 hPa averaged in a zone stretching from Mexico eastward to the Philippines (0°–20°N, 120°W– 16 120°E) (Wang et al., 2013a) in the CMIP6 historical simulation and five SSPs. One realization is 17 averaged from each model. Anomalies are shown relative to the present-day (1995–2014) mean. The 18 curves show averages over the simulations, the shadings around the SSP1-2.6 and SSP5-8.5 curves show 19 5–95% ranges, and the numbers near the top show the number of model simulations used. Further details 20 on data sources and processing are available in the chapter data table (Table 4.SM.1). 21 22 [END FIGURE 4.14 HERE] 23 24 25 In AR5, there was no specific assessment on global monsoon changes in the near term, but information can 26 be derived from CMIP5 projections of the spatial patterns of precipitation change. While the basic pattern of 27 wet regions including global monsoon regions tending to get wetter and dry regions tending to get drier is 28 apparent, large response uncertainty is evident in the substantial spread in the magnitude of projected change 29 simulated by different simulations. Over the global land monsoon regions, model uncertainty and internal 30 variability together explain 99.7% of the fraction of total variance (Zhou et al. 2020), near-term projected 31 multi-model mean precipitation changes are almost everywhere smaller than the estimated standard deviation 32 of internal variability (Figure 4.13). 33 34 The global land monsoon precipitation index, defined as the area-weighted precipitation rate in the global 35 land monsoon domain, tends to increase in the near term under all five core SSPs (Figure 4.14a) (Chen et al., 36 2020), but changes are small compared to the intermodel spread in the historical period. The Northern 37 Hemisphere summer monsoon circulation index, defined as the vertical shear of zonal winds between 850 38 and 200 hPa averaged in a zone stretching from Mexico eastward to the Philippines (0°–20°N, 120°W– 39 120°E), tends to decrease under four of the five SSP scenarios (Figure 4.14b), potentially offsetting monsoon 40 precipitation increase. Projected changes in the global monsoon circulation are also uncertain influenced by 41 internal variability and structural differences across models. In the near-term, for CMIP6 projections (Figure 42 4.14a), the multi-model mean (5–95% range) of global land monsoon precipitation change is 1.9% (–0.4– 43 4.9%), 1.6% (–1.0–5.2%), 1.3% (–1.7–3.7%), and 1.9% (–0.8–5.2%) under SSP1-2.6, SSP2-4.5, SSP3-7.0, 44 and SSP5-8.5, respectively. 45 46 In summary, we assess that near-term changes in global monsoon precipitation and circulation will be 47 affected by the combined effects of model uncertainty and internal variability, which together are larger than 48 the forced signal (medium confidence). 49 50 51 4.4.2 Cryosphere, Ocean, and Biosphere 52 53 4.4.2.1 Arctic Sea Ice 54 55 AR5 assessed that for RCP8.5, Arctic sea-ice coverage in September will drop below 1 million km2, or 56 become practically ice-free, at some point between 2040 and 2060 (Collins et al., 2013). Since AR5, there Do Not Cite, Quote or Distribute 4-42 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 has been substantial progress in understanding the response of Arctic sea ice to near-term changes in external 2 forcing. In particular, it is very likely that different trajectories of the near-term evolution of anthropogenic 3 forcing cause distinctly different likelihood ranges for very low sea-ice coverage to occur over the next two 4 decades (Notz and Stroeve, 2018). For example, there is an unlikely drop of September Arctic sea-ice 5 coverage to below 1 million km2 before 2040 for RCP 2.6, and a likely drop of September Arctic sea-ice 6 coverage to below 1 million km2 before 2040 for RCP 8.5 (medium confidence given the single study). The 7 much higher likelihood of a practically sea-ice free Arctic Ocean during summer before 2040 in RCP8.5 8 compared to RCP2.6 is consistent with related studies assessed in SROCC that find a substantially increased 9 likelihood of an ice-free Arctic Ocean for 2.0°C compared to 1.5°C mean global warming relative to pre- 10 industrial levels (Screen and Williamson, 2017; Jahn, 2018; Niederdrenk and Notz, 2018; Notz and Stroeve, 11 2018; Sigmond et al., 2018; Olson et al., 2019). 12 13 Based on results from CMIP6 models, we conclude that Arctic SIA will decrease in September in the near 14 term (Figure 4.15, high confidence). In the case of 10-year trends ending in the near term, 79% of the 15 simulations considered across all of the core SSPs project decreasing Arctic sea-ice area in September. Due 16 to less of an influence from internal variability, this number rises to 98% in the case of 30-year trends. A 17 more detailed assessment of near-term Arctic sea-ice changes can be found in Chapter 9, Section 9.3.1. A 18 detailed assessment of Antarctic sea ice changes is in Chapter 9, Section 9.3.2. 19 20 21 [START FIGURE 4.15 HERE] 22 23 Figure 4.15: CMIP6 linear trends in September Arctic sea-ice area for 10-year, 20-year, and 30-year periods 24 ending in 2021–2040 following five SSPs. Plotted are the 5–95% ranges across the ensembles of 25 simulations. The numbers at the top of the plot are the number of model simulations in each SSP 26 ensemble. The numbers near the bottom of the plot indicate the percentage of simulations across all the 27 SSPs with decreasing sea-ice area. Results are from concentration-driven simulations. Further details on 28 data sources and processing are available in the chapter data table (Table 4.SM.1). 29 30 [END FIGURE 4.15 HERE] 31 32 33 4.4.2.2 Ocean and Land Carbon flux 34 35 Ocean carbon flux is both a key feature of the physical ocean in mitigating the rise of atmospheric CO2 and a 36 driver of changes in the ocean biosphere, including changes in ocean acidity. Based on results from CMIP6 37 models, we conclude that SSP2-4.5, SSP3-7.0, and SSP5-8.5 all clearly lead to increasing 10-, 20-, and 30- 38 year trends in ocean carbon flux over the near term (Figure 4.16, high confidence). Increasing trends in ocean 39 carbon flux are less obvious in the lower-emission scenarios. Ensemble-mean trends in land carbon flux over 40 the near term are generally increasing, but these are unlikely to be detected given a large component of 41 terrestrial variability combined with model uncertainty. A more detailed assessment is in Chapter 5 Section 42 5.2.1. 43 44 In summary, it is likely that ocean carbon flux will increase in the near term under the higher emission 45 scenarios, while a large component of terrestrial variability makes it is unlikely that an increase in land 46 carbon flux will be detected over this period. 47 48 49 [START FIGURE 4.16 HERE] 50 51 Figure 4.16: CMIP6 trends in ocean and land carbon flux for 10-year, 20-year, and 30-year periods ending in 52 2021–2040. (a) Ocean carbon flux. (b) Land carbon flux. Plotted are the 5–95% ranges across the 53 ensembles of simulations, for five SSPs. The numbers at the top of the plots are the number of model 54 simulations in each SSP ensemble. Unites are PgC yr-1 per decade. Further details on data sources and 55 processing are available in the chapter data table (Table 4.SM.1). 56 Do Not Cite, Quote or Distribute 4-43 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 [END FIGURE 4.16 HERE] 2 3 4 4.4.3 Modes of Variability 5 6 This subsection assesses the near-term evolution of the large-scale modes of climate variability. Assessment 7 of the physical mechanisms and the individual feedbacks involved in the future change of each mode and 8 their role on future regional climate variability are provided in Sections 8.4.2, 9.2.3, 10.1.3 and Cross- 9 Chapter Box 10.1. 10 11 12 4.4.3.1 Northern and Southern Annular Modes 13 14 The Northern Annular Mode 15 AR5 assessed from CMIP5 simulations that there is only medium confidence in near-term projections of a 16 northward shift of NH storm track and westerlies, and an associated increase in the NAM index, because of 17 the large response uncertainty and the potentially large influence of internal variability. A tendency in the 18 near term towards a slightly more positive NAM in the three highest emission scenarios during boreal fall, 19 winter, and spring is apparent in Figure 4.17a. However, in general the projected near-term multi-model 20 mean change in the NAM is small in magnitude compared to the inter-model and/or multi-realization 21 variability within the ensemble (Figure 4.17a and Deser et al., 2012; Barnes and Polvani, 2015; Deser et al., 22 2017). 23 24 25 [START FIGURE 4.17 HERE] 26 27 Figure 4.17: CMIP6 Annular Mode index change (hPa) from 1995–2014 to 2021–2040. (a) NAM and (b) SAM. 28 The NAM is defined as the difference in zonal mean sea-level pressure (SLP) at 35°N and 65°N (Li and 29 Wang, 2003) and the SAM as the difference in zonal mean SLP at 40°S and 65°S (Gong and Wang, 30 1999). The shadings are the 5–95% ranges across the simulations. The numbers near the top of each panel 31 are the numbers of model simulations in each SSP ensemble. Further details on data sources and 32 processing are available in the chapter data table (Table 4.SM.1). 33 34 [END FIGURE 4.17 HERE] 35 36 37 On seasonal to interannual timescales, there is new evidence since AR5 that initialized predictions show 38 lower potential predictability for the boreal winter NAO than the correlation skill with respect to 39 observations (Eade et al., 2014; Baker et al., 2018; Scaife and Smith, 2018; Athanasiadis et al., 2020). This 40 has been referred to in the literature as a ‘signal-to-noise paradox’ and means that very large ensembles of 41 predictions are needed to isolate the predictable component of the NAO. While the processes that contribute 42 to predictability of the winter NAO on seasonal timescales may be distinct from the processes that drive 43 multi-decadal trends, there is emerging evidence that initialized predictions also underrepresent the 44 predictability of the winter NAO on decadal timescales (Smith et al., 2019b). Post-processing and 45 aggregation of initialized predictions may therefore reveal significant skill for predicting the winter NAO on 46 decadal timescales (Smith et al., 2020). Considering these new results since AR5, in the near-term it is likely 47 that any anthropogenic forced signal in the NAM will be of comparable magnitude or smaller than natural 48 internal variability in the NAM (medium confidence). 49 50 The Southern Annular Mode 51 AR5 assessed that it is likely that increases in GHGs and the projected recovery of the Antarctic ozone hole 52 will be the principal drivers of future SAM trends and that the positive trend in austral summer/autumn SAM 53 observed over the past several decades (AR5 Chapter 2, Section 2.4.1.2) is likely to weaken considerably as 54 stratospheric ozone recovers through to the mid-21st century. The effects of ozone depletion and recovery on 55 the SH circulation primarily occur in austral summer, while GHGs influence the SH circulation year round Do Not Cite, Quote or Distribute 4-44 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 (Gillett and Fyfe, 2013; Grise and Polvani, 2014b) and are therefore likely to be the dominant driver of 2 projected circulation changes outside of austral summer (Gillett and Fyfe, 2013; Barnes et al., 2014; 3 Solomon and Polvani, 2016). Based on current scenarios specifying future atmospheric decline of ozone 4 depleting substances (World Meteorological Organization, 2011), chemistry-climate models project the 5 Antarctic ozone hole in October to recover by around 2060 (Dhomse et al., 2018)(World Meteorological 6 Organization, 2014)(WMO, 2018). Observational evidence since AR5 shows the onset of Antarctic ozone 7 hole recovery (Solomon et al., 2016; WMO, 2018) that has been attributed to a pause in the summer SAM 8 trend over the past couple of decades (Saggioro and Shepherd, 2019; Banerjee et al., 2020). In austral 9 summer, ozone recovery and increasing GHGs will have opposing effects on the SAM over the next several 10 decades (Barnes et al., 2014). 11 12 Since AR5, there have been advances in understanding the role of internal climate variability for projected 13 near-term SH circulation trends (Solomon and Polvani, 2016). A large initial-condition ensemble following 14 the RCP4.5 emission scenario shows a monotonic positive SAM trend in austral winter. In austral summer, 15 the SAM trend over the first half of the 21st century is weaker compared to the strongly positive trend 16 observed and simulated over the late 20th century. In that model, the number of realizations required to 17 identify a detectable change in decadal mean austral winter SAM index from a year 2000 reference state 18 decreased to below five by around 2025–2030 (Solomon and Polvani, 2016). However, in DJF the same 19 criterion is not met until the second half of the 21st century, owing to the near-term opposing effects of 20 ozone recovery and GHGs on the austral-summer SAM. In austral summer, forced changes in the SAM 21 index in the near-term are therefore likely to be smaller than changes due to internal variability (Figure 22 4.17:b; Barnes et al., 2014; Solomon and Polvani, 2016). 23 24 CMIP6 models show a tendency in the near-term towards a more positive SAM index especially in the 25 austral winter (JJA; Figure 4.17b). In all seasons, the differences between the central estimates of the change 26 in the SAM index for each SSP are much smaller than the inter-model ensemble spread. The number of 27 CMIP6 realizations in Figure 4.17b is larger than the suggested threshold of five realizations needed to 28 detect a significant near-term change in decadal-mean austral winter SAM index for a single CMIP5 model 29 (Solomon and Polvani, 2016), and yet the 5–95% intervals on the CMIP6 ensemble spread encompass zero 30 for all core SSPs. This suggests both internal variability and model uncertainty contribute to the CMIP6 31 ensemble spread in near-term SAM index changes. Based on these results, it is more likely than not that in 32 the near-term under all assessed SSP scenarios the SAM index would become more positive than in present- 33 day in austral autumn, winter and spring. 34 35 An influence of forcing agents other than stratospheric ozone and GHGs, such as anthropogenic aerosols, on 36 SAM changes over the historical period has been reported in some climate models (Rotstayn, 2013), but the 37 response across a larger set of CMIP5 models is not robust (Steptoe et al., 2016) and depends on how 38 tropospheric temperature response to aerosols (Choi et al., 2019). This gives low confidence in the potential 39 influence of anthropogenic aerosols on the SAM in the future. 40 41 42 4.4.3.2 El Niño-Southern Oscillation 43 44 AR5 assessed that it is very likely that the ENSO will remain the dominant mode of interannual variability in 45 the future but did not specify its change in near term. A subset of CMIP5 models that simulate the ENSO 46 Bjerknes index most realistically show an increase of ENSO SST amplitude in the near-term future and 47 decline thereafter (Kim et al., 2014). However, detection of robust near-term changes of ENSO SST 48 variability in response to anthropogenic forcing is difficult to achieve due to pronounced unforced low- 49 frequency modulations of ENSO (Wittenberg, 2009; Maher et al., 2018; Wengel et al., 2018). Figure 4.10 in 50 Section 4.3.3.2 using CMIP6 models also shows no robust change in ENSO SST variability in the near term. 51 52 While there is no strong model consensus on the change in amplitude of ENSO SST variability, the 53 amplitude of ENSO-associated rainfall variability likely increases (Power et al., 2013; Cai et al., 2015). 54 Analysis of CMIP6 models shows a slight increasing trend in amplitude of rainfall variability over Niño3.4 55 region in the near term attributable to mean moisture increase, regardless of changes in ENSO SST Do Not Cite, Quote or Distribute 4-45 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 variability (Figure 4.10) However, there are no distinguishable changes in the rainfall variability among four 2 SSPs with significant model spread in the near term. Hence, no robust change in amplitude of ENSO SST 3 and rainfall variability is expected in the near term although the rainfall variability slightly increases 4 (medium confidence). 5 6 7 4.4.3.3 Indian Ocean Basin and Dipole Modes. 8 9 Important modes of interannual climate variability with pronounced climate impacts in the Africa-Indo- 10 Pacific areas of the globe are the Indian Ocean Dipole (IOD), which is closely related to- and often coincides 11 with ENSO phases (Stuecker et al., 2017), and the Indian Ocean Basin (IOB) mode, which is often described 12 as a capacitor effect in response to ENSO (Xie et al., 2009; Du et al., 2013) and can feed back onto ENSO 13 evolution (Cai et al., 2019a). IOD and IOB are extensively described in Annex IV, Section AIV2.4. 14 15 The projected climate mean state changes in the tropical Indian Ocean resemble a positive IOD state, with 16 faster warming in the west compared to the east. This mean state change will potentially lead to a reduction 17 in the amplitude of IOD events, albeit with no robust change in IOD frequency (Cai et al., 2014b). There is 18 no robust evidence yet suggesting a cessation of IOD variability or a significant change in the IOB mode in 19 the near-term. 20 21 22 4.4.3.4 Tropical Atlantic Modes 23 24 Interannual variability of the tropical Atlantic can be described in terms of two main climate modes: the 25 Atlantic Equatorial Mode and the Atlantic Meridional Mode (AMM) (Annex IV, Section AIV2.5). The 26 Atlantic Equatorial Mode, also commonly referred to as the Atlantic Niño or Atlantic Zonal Mode, is 27 associated with SST anomalies near the equator, peaking in the eastern basin, while the AMM is 28 characterized by an inter-hemispheric gradient of SST and wind anomalies. Both modes are associated with 29 changes in the ITCZ and related winds and exert a strong influence on the climate in adjacent and remote 30 regions. 31 32 Despite considerable improvements in CMIP5 with respect to CMIP3, most CMIP5 models have difficulties 33 in simulating the mean climate of the tropical Atlantic (Mohino et al., 2019) and are not able to correctly 34 simulate the main aspects of Tropical Atlantic Variability (TAV) and associated impacts. This is presumably 35 the main reason why there is a lack of specific studies dealing with near-term changes in tropical Atlantic 36 modes. Nevertheless, AR5 reported that the ocean is more predictable than continental areas at the decadal 37 timescale (Kirtman et al., 2013). In particular, the predictability in the tropical oceans is mainly associated 38 with decadal variations of the external forcing component. Since the AMV affects the tropical Atlantic, near- 39 term variations of the AMV can modulate the Equatorial Mode and the AMM as well as associated impacts. 40 41 There are no specific studies focusing on near-term changes in tropical Atlantic modes; nevertheless, decadal 42 predictions show that although the North Atlantic stands out in most CMIP5 models as the primary region 43 where skill might be improved because of initialization, encouraging results have also been found in the 44 tropical Atlantic (Meehl et al., 2014). The effect of initialization in the tropical Atlantic is not only visible in 45 surface temperature but also in the subsurface ocean (Corti et al., 2015). In particular, initialization improves 46 the skill via remote ocean conditions in the North Atlantic subpolar gyre and tropical Pacific, which 47 influence the tropical Atlantic through atmospheric teleconnections (Dunstone et al., 2011; Vecchi et al., 48 2014; García-Serrano et al., 2015a). Improvements of some aspects of climate prediction systems 49 (initialization techniques, large ensembles, increasing model resolution) have also led to skill improvements 50 over the tropical Atlantic (Pohlmann et al., 2013; Monerie et al., 2017; Yeager and Robson, 2017). 51 52 Recent studies have shown that the AMV can modulate not only the characteristics of the Atlantic Niños, but 53 also their inter-basin teleconnections (Indian and Pacific). In particular, the Atlantic Niño–ENSO 54 relationship is strongest during negative AMV phases (Martín-Rey et al., 2014; Losada and Rodríguez- 55 Fonseca, 2016) when equatorial Atlantic SST variability is enhanced (Martín-Rey et al., 2017; Lübbecke et Do Not Cite, Quote or Distribute 4-46 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 al., 2018). 2 3 Based on CMIP5 and available CMIP6 results, we conclude that there is a lack of studies on the near-term 4 evolution of TAV and associated teleconnections for a comprehensive assessment. However, some studies 5 show that despite severe model biases there are skilful predictions in the mean state of tropical Atlantic 6 surface temperature several years ahead (medium confidence), though skill in simulated variability has not 7 been assessed yet. Decadal changes in the Atlantic Niño spatial configuration and associated teleconnections 8 might be modulated by the AMV, but there is limited evidence and therefore low confidence in these results. 9 10 11 4.4.3.5 Pacific Decadal Variability 12 13 Climate variability of the Pacific Ocean on decadal and interdecadal timescales is described in terms of a 14 number of quasi-oscillatory SST patterns such as the Pacific Decadal Oscillation (PDO) (Mantua et al., 15 1997) and the Interdecadal Pacific Oscillation (IPO) (Folland, 2002), which are referred to as the Pacific 16 Decadal Variability (PDV) (Newman et al., 2016). PDV comprises an inter-hemispheric pattern that varies at 17 decadal-to-interdecadal timescales (see Chapter 3, Figure 3.35). However, although the spatial domains to 18 derive the IPO and PDO indices differ, and despite uncertainty related to trend removal and time-filtering 19 (Newman et al., 2016; Tung et al., 2019), the IPO and PDO are highly correlated in time and they will be 20 assessed together as the PDV (see Annex IV, Section AIV.2.6). 21 22 AR5 assessed that near-term predictions of PDV (then referred to as PDO or IPO) were largely model 23 dependent (Mochizuki et al., 2012; van Oldenborgh et al., 2012), not robust to sampling of initialization 24 start-dates, overall not statistically significant, and worse than persistence (Doblas-Reyes et al., 2013), 25 although some studies showed positive skill for PDV (Mochizuki et al., 2010; Chikamoto et al., 2013). The 26 CMIP5 decadal-prediction ensemble yielded no prediction skill of SST over the key PDV centres of action in 27 the Pacific Ocean, both at 2–5 year and 6–9 year forecast averages (Doblas-Reyes et al., 2013; Guemas et al., 28 2013; Boer and Sospedra-Alfonso, 2019). 29 30 Since AR5, the processes causing the multi-decadal variability in the Pacific Ocean have become better 31 understood (Newman et al., 2016)(Henley, 2017). However, the relative importance of tropical and 32 extratropical processes underlying PDV remains unclear; although it seems to be stochastically driven rather 33 than self-excited (Liu, 2012; Liu and Di Lorenzo, 2018), the South Pacific being a key region for the tropical 34 branch of PDV (Chung et al., 2019; Liguori and Di Lorenzo, 2019). 35 36 Because PDV represents not one, but many dynamical processes, it represents a challenge as a target for 37 near-term climate predictions and projections. The new generation of decadal forecast systems keeps 38 showing poor (Shaffrey et al., 2017) to moderate (Smith et al., 2019b) multi-year prediction skill for PDV, 39 although the potential for forecasting capabilities is demonstrated in case studies (Meehl and Teng, 2012; 40 Meehl et al., 2014). For the near-term, a transition of PDV from the negative phase (1999–2012) towards a 41 positive phase is predicted in the coming years (2013-2022) (Meehl et al., 2016). 42 43 The PDV has been shown to influence the pace of global warming (see Cross Chapter Box 3.1), but the 44 extent to which PDV is externally forced or internally generated (Mann et al., 2020) remains an open 45 question, and there is still no robust evidence. Thus, there is low confidence on how the PDV will evolve in 46 the near-term (Bordbar et al., 2019). 47 48 49 4.4.3.6 Atlantic Multidecadal Variability 50 51 The Atlantic Multi-decadal Variability (AMV) is a large-scale climate mode accounting for the main 52 fluctuations in North Atlantic SST on multi-decadal timescales (Annex IV, Section AIV.2.7). The AMV 53 influences air temperatures and precipitation over adjacent and remote continents, and its undulations can 54 partially explain the observed variations in the NH mean temperatures (Steinman et al., 2015). The origin of 55 this variability is still uncertain. Ocean dynamics (e.g., changes in the AMOC), external forcing, and local Do Not Cite, Quote or Distribute 4-47 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 atmospheric forcing all seem to play a role (Menary et al., 2015; Ruprich-Robert and Cassou, 2015; Brown 2 et al., 2016; Cassou et al., 2018; Wills et al., 2019). Recent studies have discussed that the ocean dynamics 3 play an active role in generating AMV (Oelsmann et al., 2020) and its interplay with the NAO (Vecchi et al., 4 2017; Zhang et al., 2019a; Kim et al., 2020), although natural and anthropogenic external forcing might be 5 crucial in modulating its amplitude and timing (Bellucci et al., 2017)(Bellomo et al., 2018; Andrews et al., 6 2020; Borchert et al., 2021)(Mann et al., 2021) (see Section 3.7.7 and Section AIV.2.7). 7 8 AR5 assessed with high confidence that initialized predictions can improve the skill for temperature over the 9 North Atlantic, in particular in the sub-polar branch of AMV, compared to the projections, for the first five 10 years (see WGI AR5 Figures 11.3 and 11.4). However, non-initialized predictions showed positive 11 correlation over the same time-range as well, consistent with the notion that part of this variability is caused 12 by external forcing (see Chapter 3 Section 3.7.7). 13 14 Since AR5, near-term initialized predictions, both multi-model (Bellucci et al., 2015a; García-Serrano et al., 15 2015b; Smith et al., 2019b) and single-model ensembles (Marotzke et al., 2016)(Simpson et al., 16 2018c)(Yeager et al., 2018)(Hermanson et al., 2020)(Bilbao et al., 2021), confirm substantial skill in 17 hindcasting North-Atlantic SST anomalies on a time range of 8–10 years. On the same time range, (Borchert 18 et al., 2021) show a substantial improvement in the prediction of the subpolar gyre SST (the northern 19 component of the AMV) in CMIP6 models compared to CMIP5, in both initialized and non-initialized 20 simulations. The higher skill can be attributed to a more accurate response in CMIP6 models to natural 21 forcing, possibly originating from the AMOC-related delayed response to volcanic eruptions, which 22 contribute to the SST variations in the subpolar gyre (Hermanson et al., 2020). 23 24 Initialization contributes to the reduction of uncertainty and to predicting subpolar SST amplitude (Borchert 25 et al., 2021). Yet, skill in predicting the AMV is not always translated into equally successful predictions of 26 temperature and precipitation over the nearby land and ocean regions (Langehaug et al., 2017). This might 27 be related to systematic model errors in the simulation of the spatial and temporal structure of the AMV and 28 too weak associated teleconnections (see Section 3.7.7), and also to the larger noise in regional land 29 variables compared to the AMV index. However, AMV predictions can be used as proxies to predict other 30 variables such as precipitation over Western Europe and Eurasia and SAT over Mediterranean, northern 31 Europe and northeast Asia (Årthun et al., 2018; Borchert et al., 2019; Ruggieri et al., 2021) whose 32 relationship with North Atlantic SST is robust in observations, but not well captured in climate models. 33 34 Encouraging results about the prediction of land precipitation linked to the warm AMV phase (see Chapter 3 35 Section 3.7.7 and Annex IV, Figure AIV.2.7) on a 2-9 year timescale are reported in the multi-model study 36 by (Smith et al., 2019b). Positive correlations with observations are found in the Sahel, South America, the 37 Maritime Continent. Analyses from large-ensemble decadal prediction systems such as the community Earth 38 system model decadal prediction large ensemble (CESM-DPLE) (Yeager et al., 2018) show an improvement 39 with respect to the CMIP5 decadal hindcasts (Martin and Thorncroft, 2014b) in forecasting Sahel 40 precipitation over three to seven years, which is consistent with the current understanding of AMV impact 41 over Africa (Mohino et al., 2016; Smith et al., 2019b). CESM-DPLE predicts drought conditions over the 42 Sahel through 2020, which is compatible with a shift towards a negative phase of AMV as a result of a 43 weakening of the AMOC, advocated by a number of studies (Hermanson et al., 2014; Robson et al., 2014; 44 Yeager et al., 2015). 45 46 In summary, the confidence in the predictions of AMV and its effects is medium. However, there is high 47 confidence that the AMV skill over 5–8-year lead times is improved by using initialized predictions, 48 compared to non-initialized simulations. 49 50 51 4.4.4 Response to Short-Lived Climate Forcers and Volcanic Eruptions 52 53 Mitigation of SLCFs affects future climate projections and could alter the time of emergence of 54 anthropogenic climate change signals. AR5 assessed that emission reductions aimed at decreasing local air 55 pollution could have a near-term warming impact on climate (high confidence) (Kirtman et al., 2013). Do Not Cite, Quote or Distribute 4-48 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Because of their shorter lifetimes, reductions in emissions of SLCF species mainly influence near-term 2 GSAT trends (Chalmers et al., 2012; Shindell et al., 2017; Shindell and Smith, 2019), but on decadal 3 timescales the near-term response to even very large reductions in SLCFs may be difficult to detect in the 4 presence of large internal climate variability (Samset et al., 2020). The changes in SLCF emissions during 5 the COVID-19 pandemic has resulted in a small net radiative forcing without a discernible impact on GSAT 6 (Cross-Chapter Box 6.1). SLCF mitigation also leads to a higher GSAT in the mid- to long-term (Smith and 7 Mizrahi, 2013; Stohl et al., 2015; Hienola et al., 2018) and can influence peak warming during the 21st 8 century (Hienola et al., 2018; Rogelj et al., 2014). This section focuses on the total effect of SLCF changes 9 on GSAT projections in the SSP scenarios. A more detailed breakdown of the separate climate effects of 10 SLCF species and precursor species can be found in Sections 6.7.2 and 6.7.3. 11 12 A model experiment based on the SSP3-7.0 scenario with aerosols, their precursors, and non-methane 13 tropospheric ozone precursors set to SSP1-1.9 abundances (SSP3-7.0-lowSLCF-highCH4; Collins et al., 14 2017) shows a projected multi-model mean GSAT anomaly that is higher by 0.22ºC at mid-century (2045- 15 2054) compared to SSP3-7.0 (Figure 4.18:; Allen et al., 2020), but this difference is smaller than the 16 intermodel spread of the SSP3-7.0 projections based on the CMIP6 models. Note the SSP3-7.0-lowSLCF- 17 highCH4 experiment does not perturb methane from SSP3-7.0 concentrations. A modified SSP3-7.0- 18 lowSLCF-lowCH4 scenario that also includes methane mitigation shows a lower GSAT by mid-century 19 compared to SSP3-7.0 (Allen et al., 2021). 20 21 22 [START FIGURE 4.18 HERE] 23 24 Figure 4.18: Influence of SLCFs on projected GSAT change. Change is shown relative to the 1995–2014 average 25 (left axis) and relative to the 1850–1900 average (right axis). The comparison is for CMIP6 models for 26 the AerChemMIP (Collins et al., 2017) SSP3-7.0-lowSLCF-highCH4 experiment (note in the original 27 experiment protocol this is called SSP3-7.0-lowNTCF), where concentrations of short-lived species are 28 reduced compared to reference SSP3-7.0 scenario. The curves show averages over the r1 simulations 29 contributed to the CMIP6 exercise, the shadings around the SSP3-7.0 curve shows 5–95% ranges and the 30 numbers near the top show the number of model simulations. Further details on data sources and 31 processing are available in the chapter data table (Table 4.SM.1). 32 33 [END FIGURE 4.18 HERE] 34 35 36 Building on CMIP6 results for the effects of reducing SLCF emissions from a baseline of SSP3-7.0, the 37 overall contribution of SLCFs to GSAT changes in the marker SSPs are now quantified using a simple 38 climate model emulator. For consistency with Section 6.7.2 and Figure 6.22, the basket of SLCF compounds 39 considered includes aerosols, ozone, methane, black carbon on snow and hydrofluorocarbons (HFCs) with 40 lifetimes of less than 50 years. In the five marker SSPs considered, the net effect of SLCFs contributes to a 41 higher GSAT in the near, mid- and long term (Table 4.6, Section 6.7.2). In the SSP1-1.9 and SSP1-2.6 42 scenarios, SLCFs contribute to a higher GSAT by a central estimate of around 0.3 °C compared to 1995– 43 2014 across the three-time horizons. In the long-term, the 0.3 °C warming due to SLCFs in SSP1-2.6 can be 44 compared to the assessed very likely GSAT change for this period of 0.5–1.5 °C (Section 4.3.4; Table 4.5). 45 The SSP2-4.5, SSP3-7.0 and SSP5-8.5 scenarios all show a larger SLCF effect on GSAT in the long term 46 relative to the near term. In SSP3-7.0, the long-term warming due to SLCFs by 0.7 °C can be compared with 47 the assessed very likely GSAT anomaly for this period of 2.0–3.7 °C (Section 4.3.4). In summary, it is very 48 likely that changes in SLCFs contribute to an overall warmer GSAT over the near, mid- and long term in the 49 five SSP scenarios considered (Table 4.6, Section 6.7.2; Figure 6.22). 50 51 52 [START TABLE 4.6 HERE] 53 54 Table 4.6: The net effect of SLCFs on GSAT change. Changes in 20-year averaged GSAT relative to 1995–2014 55 for 2021–2040, 2041–2060, and 2081–2100 for the five marker SSP scenarios. Values give the median 56 and, in parentheses, the 5–95% range calculated from a 2,237-member ensemble of the two-layer Do Not Cite, Quote or Distribute 4-49 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 emulator that is driven with the ERF projections, including uncertainties, described in Chapter 7 2 Supplementary Material 7.SM.1.4. The ensemble is constrained to assessed ranges of ECS, TCR, ocean 3 heat content change, GSAT response, and carbon cycle metrics (Section 7.3.5; Chapter 7 Supplementary 4 Material 7.SM.2.2). The GSAT contribution of individual forcer responses use the difference between 5 parallel runs of the constrained two-layer model with all anthropogenic forcing and all anthropogenic 6 forcing with the component of interest (e.g. methane) removed (Chapter 7 Supplementary Material 7 7.SM.2.3). Values are given to 1 decimal place. 8 Units = °C SSP1-1.9 SSP1-2.6 SSP2-4.5 SSP3-7.0 SSP5-8.5 Near term 0.2 (0.1, 0.3) 0.2 (0.1, 0.3) 0.2 (0.1, 0.3) 0.2 (0.1, 0.3) 0.3 (0.2, 0.4) (2021–2040) Mid-term 0.2 (0.0, 0.4) 0.2 (0.0, 0.4) 0.3 (0.2, 0.4) 0.3 (0.2, 0.4) 0.5 (0.3, 0.7) (2041–2060) Long term 0.1 (-0.1, 0.4) 0.2 (0.0, 0.4) 0.3 (0.1, 0.6) 0.5 (0.4, 0.8) 0.7 (0.4, 1.0) (2081–2100) 9 10 [END TABLE 4.6 HERE] 11 12 13 In addition to effects on GSAT, SLCFs affect other aspects of the global climate system (Section 6.7.2). The 14 additional warming at high northern latitudes associated with projected reductions in aerosol emissions over 15 the 21st century leads to a more rapid reduction in Arctic sea-ice extent in the RCP scenarios (Gagné et al., 16 2015). Furthermore, mitigation of non-methane SLCFs in the SSP3-7.0-lowSLCF-highCH4 scenario causes 17 an increase in global mean precipitation, with larger regional changes in southern and eastern Asia (Allen et 18 al., 2020). 19 20 The main uncertainties in climate effects of SLCFs in the future come from: (i) the uncertainty in 21 anthropogenic aerosol ERF (Section 7.3.3); (ii) uncertainty in the relative emissions of different SLCFs that 22 have warming and cooling effects in the current climate (Section 6.2); and (iii) physical uncertainty 23 including the efficacy of the climate response to SLCFs compared to long-lived GHGs (Marvel et al., 2016; 24 Richardson et al., 2019). One example of physical uncertainty is that the shortwave radiative forcing from 25 methane was neglected in previous calculations (Etminan et al., 2016; Collins et al., 2018), which affects 26 understanding of present-day and future methane ERF (Modak et al., 2018b). Another example of physical 27 uncertainty is projected changes in lightning-NOx production, which contributes to future ozone radiative 28 forcing (Banerjee et al., 2014, 2018; Finney et al., 2018). 29 30 Another factor that could substantially alter projections in the near-term would be the occurrence of a large 31 explosive volcanic eruption, or even a decadal to multi-decadal sequence of small-to-moderate volcanic 32 eruptions as witnessed over the early 21st century (Santer et al., 2014; Cross-Chapter Box 4.1). An eruption 33 similar to the last large tropical eruption, Mount Pinatubo in the Philippines in June 1991, is expected to 34 cause substantial Northern Hemisphere (NH) cooling, peaking between 0.09°C and 0.38°C and lasting for 35 three to five years, as indicated by climate model simulations over the past millennium (e.g., Jungclaus et al., 36 2010). Phase 3 of Paleoclimate Modelling Intercomparison Project (PMIP3) simulated a significant NH 37 cooling in response to individual volcanic events (peaks between 0.1°C and 0.5°C, depending on model, 38 during the first year after the eruption) that lasts for three to five years. On a regional scale, the double 39 volcanic events that occurred in 536 and 540 CE resulted in a cooling of 2°C (Büntgen et al., 2016a; Toohey 40 et al., 2016a). 41 42 Since AR5, there has been growing progress in understanding the climate impacts of volcanic eruptions. 43 Volcanic forcing is regarded as the dominant driver of forced variability in preindustrial surface air 44 temperature (Schurer et al., 2013a, 2014). Large eruptions in the tropics and high latitudes were primary 45 drivers of interannual-to-decadal temperature variability in the Northern Hemisphere during the past 2,500 46 years, with cooling persisting for up to ten years after some of the largest eruptive episodes (Sigl et al., 47 2015). Repeated clusters of volcanic eruptions can induce a net negative radiative forcing that results in a 48 centennial and global scale cooling trend via a decline in mixed-layer oceanic heat content (McGregor et al., Do Not Cite, Quote or Distribute 4-50 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2015). The response to multi-decadal changes in volcanic forcing (representing clusters of eruptions) shows 2 similar cooling in both simulations and reconstructions of NH temperature. Volcanic eruptions generally 3 result in decreased global precipitation for up to a few years following the eruption (Iles and Hegerl, 2014, 4 2015; Man et al., 2014), with climatologically wet regions drying and climatologically dry regions wetting 5 (medium confidence), which is opposite to the response under global warming (Held and Soden, 2006b; Iles 6 et al., 2013; Zuo et al., 2019a, 2019b). El Niño-like warming appears after large volcanic eruptions that is 7 seen in both observations (Adams et al., 2003; McGregor et al., 2010a; Khodri et al., 2017) and climate 8 model simulations (Ohba et al., 2013; Pausata et al., 2015; Colose et al., 2016; Stevenson et al., 2016; Khodri 9 et al., 2017; Predybaylo et al., 2017; Zuo et al., 2018). The large tropical eruptions are coincident with 10 positive Indian Ocean Dipole events (Maher et al., 2015). 11 12 In AR5, uncertainty due to future volcanic activity was not considered in the assessment of the CMIP5 21st 13 century climate projections (Taylor et al., 2012; O’Neill et al., 2016). Since AR5, there has been 14 considerable progress in quantifying the impacts of volcanic eruptions on decadal climate prediction and 15 longer-term climate projections (Meehl et al., 2015; Swingedouw et al., 2015, 2017; Timmreck et al., 2016; 16 Bethke et al., 2017; Illing et al., 2018). By exploring 60 possible volcanic futures under RCP4.5, it has been 17 demonstrated that the inclusion of time-varying volcanic forcing may enhance climate variability on annual- 18 to-decadal timescales (Bethke et al., 2017). Consistent with a tropospheric cooling response, the change in 19 ensemble spread in the volcanic cases is skewed towards lower GSAT relative to the non-volcanic cases 20 (Cross-Chapter Box 4.1 Figure 1). In these simulations with multiple volcanic forcing futures there is: 1) an 21 increase in the frequency of extremely cold individual years; 2) an increased likelihood of decades with 22 negative GSAT trend (decades with negative GSAT trends become 50% more commonplace); 3) later 23 anthropogenic signal emergence (the mean time at which the signal of global warming emerges from the 24 noise of natural climate variability is delayed almost everywhere) (high confidence); and 4) a 10% overall 25 reduction in global land monsoon precipitation and a 20% overall increase in the ensemble spread (Man et 26 al., 2021). 27 28 [START Cross-Chapter Box 4.1 HERE] 29 30 Cross-Chapter Box 4.1: The climate effects of volcanic eruption 31 32 Contributing Authors: Sarah Connors (France/UK), Amanda Maycock (UK), Peter Thorne (Ireland/UK), 33 Nicolas Bellouin (UK/ France), Ingo Bethke (Norway), Deliang Chen (Sweden), Annalisa Cherchi (Italy), 34 Alejandro Di Luca (Australia/Argentina), Piers Forster (UK), Nathan Gillett (Canada), Darrell Kaufmann 35 (USA), June-Yi Lee (Republic of Korea), Elizaveta Malinina (Canada), Seung-Ki Min (Republic of Korea), 36 Johannes Quaas (Germany), Alex C. Ruane (USA), Jean-Baptiste Sallée (France), Sonia I. Seneviratne 37 (Switzerland), Chris Smith (UK), , Matthew Toohey (Canada), Andrew Turner (UK), Cunde Xiao (China), 38 Tianjun Zhou (China). 39 40 Before the industrial period, explosive volcanic eruptions were the largest source of forced climate 41 variability globally on interannual to centennial timescales (Section 2.2). While usually omitted from 42 scenarios used for future climate projections, as they are unpredictable, volcanic eruptions have the potential 43 to influence future climate on multi-annual to decadal timescales and affect many climatic impact drivers (as 44 defined in Sections 12.1, 12.3). Since AR5, more comprehensive paleo evidence and observations, as well as 45 improved modelling have advanced understanding of the climate response to past volcanic eruptions. 46 Building on multiple chapter assessments, this box synthesizes how volcanic eruptions affect climate and 47 considers implications of possible future events. 48 49 How frequent are volcanic eruptions? 50 51 Proxy records show that large volcanic eruptions with effective radiative forcing (ERF) more negative than – 52 1 W m-2 occurred on average twice a century throughout the last 2500 years, the most recent being Pinatubo 53 in 1991 (Section 2.2.2). About eight larger eruptions (ERF stronger than –5 Wm-2) also occurred during this 54 period (Figure 2.2), notably Tambora~1815 and Samalas~1257. A Samalas-type eruption may occur 1–2 55 times per millennium on average (Newhall et al., 2018). Typically, three in every four centuries have Do Not Cite, Quote or Distribute 4-51 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 experienced at least one eruption stronger than –1 W m-2 (Pinatubo or larger). The volcanic aerosol burden 2 was 14% lower during the 20th century compared to the average of the preceding 24 centuries (Section 3 2.2.2), whereas the 13th century was among the most volcanically active, with four eruptions exceeding that 4 of Pinatubo-1991 (Sigl et al., 2015). 5 6 Past climate responses to volcanic activity 7 8 Major eruptions drive a range of climate system responses for several years depending upon whether the 9 eruption occurs in the tropics (stratospheric aerosol dispersion into both hemispheres) or the extra-tropics 10 (dispersion into the hemisphere of eruption) owing to the Brewer-Dobson circulation. The climatic response 11 also depends on the effective injection height, sulphur mass injected, and time of year of the eruption 12 (Marshall et al., 2019, 2020). These factors determine the total mass, lifetime and optical properties of 13 volcanic aerosol in the stratosphere and influence the stratospheric aerosol optical depth (sAOD). The ERF 14 from volcanic stratospheric aerosol is assessed to be –20 ± 5 W m–2 per unit sAOD (Section 7.3.4.6). 15 16 Due to the direct radiative effect of volcanic stratospheric aerosols, large volcanic eruptions lead to an 17 overall decrease of GSAT, which can extend to multi-decadal or century timescales in the case of clustered 18 volcanism (Section 3.3.1.1, Schurer et al., 2013; McGregor et al., 2015; Sigl et al., 2015; Kobashi et al., 19 2017; Zambri et al., 2017; Brönnimann et al., 2019; Neukom et al., 2019). Large eruptions also increase the 20 frequency of extremely cold individual years and the likelihood of cooling trends occurring in individual 21 decades (Cross-chapter Box 3.1, Section 4.4.4, Paik and Min, 2018). Re-dating of ice core chronologies now 22 confirms that the coldest decades of the past ~2000 years are the outcome of volcanic eruptions (Sigl et al., 23 2015; Büntgen et al., 2016b; Toohey et al., 2016b; Neukom et al., 2019). CMIP5 and CMIP6 models 24 reproduce the decreased GSAT that follows periods of intense volcanism. New reconciliations between 25 simulations and proxy-based reconstructions of past eruptions have been achieved through better Earth 26 System Model representation of volcanic plume chemical compositions (Legrande et al., 2016; Marshall et 27 al., 2020; Zhu et al., 2020a). Yet, remaining disagreements reflect differences in the volcanic forcing datasets 28 used in the simulations (medium confidence) (Section 3.3.1.1, Figure 3.2c). 29 30 Although incomplete, proxy records show large impacts upon contemporary society from eruptions such as 31 1257 Samalas and 1815 Tambora, the latter resulting in ‘the year without a summer’ with multiple harvest 32 failures across the NH (e.g., Raible et al., 2016). Comparing CMIP5 multi-model simulations with 33 observations has improved understanding of the hydrological responses to 20th-century eruptions, 34 particularly global land monsoon drying, and associated uncertainties (Section 3.3.2.2). Global-mean land 35 precipitation decreases for up to a few years following the eruption, with climatologically wet regions 36 drying and dry regions wetting (Section 3.3.2.2, Section 4.4.4). Changes in monsoon circulations occur with 37 a general weakening of tropical precipitation (Section 8.5.2.3) and a decrease in extreme precipitation over 38 global monsoon regions (Section 11.4.4). Monsoon precipitation in one hemisphere tends to be enhanced by 39 eruptions occurring in the other hemisphere or reduced if they occur in the same hemisphere (Section 40 3.3.2.2; 8.5.2.3). Volcanic eruptions have been linked to the onset of El Niño followed by La Niña although 41 this connection remains contentious (Adams et al., 2003; Bradley et al., 2003; McGregor et al., 2010b; 42 Khodri et al., 2017; Liu et al., 2018a; Sun et al., 2019; Paik et al., 2020; Predybaylo et al., 2020). Volcanic 43 activity could drive short-term (1-3 year) positive changes in the annual SAM index through modulations in 44 the extratropical temperature gradient and wave driving of the polar stratosphere (Yang and Xiao, 2018). In 45 the cryosphere, Arctic sea-ice extent increases for years to decades (Gagné et al., 2017b), and modelling 46 indicates that sea-ice/ocean feedbacks can prolong cooling long after volcanic aerosols are removed (Miller 47 et al., 2012). On annual timescales, the ocean buffers the atmospheric response to volcanic eruptions by 48 storing the cooling in the ocean subsurface, then feeding it back to the atmosphere. Large eruptions affect 49 ocean heat content and thermosteric sea level over decadal-to-centennial scales (Section 9.2.2.1). 50 51 Potential implications on 21st century projections 52 53 Given the unpredictability of individual eruptions, volcanic forcing is prescribed as a constant background 54 loading in CMIP6 models (Eyring et al., 2016). This means the effects of potential large volcanic eruptions 55 are largely absent from model projections, and few studies have addressed the potential implications on 21st Do Not Cite, Quote or Distribute 4-52 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 century warming. One study considered future scenarios with hypothetical volcanic eruptions consistent with 2 levels of CE volcanic activity (Bethke et al., 2017) under RCP4.5 and found that climate projections could be 3 substantially altered (Cross Chapter Box 4.1 Figure 1). Although temporary, close to pre-industrial level 4 temperatures could be experienced globally for a few years after a 1257 Samalas-sized eruption. Several 5 other key climate indicators are also changed substantially, consistent with evidence from past events. 6 (Bethke et al., 2017) suggest that an eruption early in the 21st century could delay the timing of crossing 7 1.5°C global warming by several years. Clustered eruptions would have substantial impact upon GSAT 8 evolution throughout the century (Cross-Chapter Box 4.1 Figure 1), and could have far-reaching 9 implications, as observed for past eruptions. For near-term response options, decadal prediction models can 10 update 21st-century projections once a volcanic eruption occurs (Timmreck et al., 2016). 11 12 Summary 13 14 It is likely that at least one large eruption will occur during the 21st century. Such an eruption would reduce 15 GSAT for several years, decrease global-mean land precipitation, alter monsoon circulation, modify extreme 16 precipitation, and change the profile of many regional climatic impact-drivers. A low likelihood high impact 17 outcome would be several large eruptions that would greatly alter the 21st century climate trajectory 18 compared to SSP-based ESM projections. 19 20 21 [START CROSS-CHAPTER BOX 4.1, FIGURE 1 HERE] 22 23 Cross-Chapter Box 4.1, Figure 1: Potential impact of volcanic eruption on future global temperature change. 24 CMIP5 projections of possible 21st-century futures under RCP4.5 after a 1257 Samalas 25 magnitude volcanic eruption in 2044, from (Bethke et al., 2017). a, Volcanic ERF of the 26 most volcanically active ensemble member, estimated from SAOD. b, Annual-mean 27 GSAT. Ensemble mean (solid) of future projections including volcanoes (blue) and 28 excluding volcanoes (red) with 5–95% range (shading) and ensemble minima/maxima 29 (dots); evolution of the most volcanically active member (black). Data created using a 30 SMILE approach with NorESM1 in its CMIP5 configuration. See Section 2.2.2 and 31 Section 4.4.4 for more details. Further details on data sources and processing are available 32 in the chapter data table (Table 4.SM.1). 33 34 [END CROSS-CHAPTER BOX 4.1, FIGURE 1 HERE] 35 36 37 [END Cross-Chapter Box 4.1 HERE] 38 39 40 4.5 Mid- to Long-term Global Climate Change 41 42 4.5.1 Atmosphere 43 44 This section assesses how the global atmospheric indicators assessed in Section 4.3 manifest themselves in 45 large-scale spatial patterns of atmospheric change in the mid-term (2041–2060) and long term (2081–2100). 46 The patterns of change in any given future period represent a combination of unforced internal variability 47 and a forced response including their interaction (Deser et al., 2016). The role of internal variability is much 48 larger at the local to regional scale than in the global mean projections. We here assess multi-model mean 49 patterns based on CMIP6 models without any weighting or emergent constraints. The mean represents an 50 estimate of the forced response and is a more homogeneous pattern than the 20-year mean change patterns in 51 any individual model realization (Knutti et al., 2010). 52 53 54 4.5.1.1 Near-Surface Air Temperature 55 56 Patterns of near-surface air temperature changes show widespread warming by 2041–2060 and 2081–2100 Do Not Cite, Quote or Distribute 4-53 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 (Figure 4.19) for all SSPs relative to 1995–2014. The area fraction experiencing warming increases with the 2 level of global mean warming. As GSAT continues to increase, it is very likely that by the middle and the 3 end of the 21st century most of the global land and ocean areas will be warmer than in 1995–2014 (high 4 confidence) (see also Section 4.3.1.1). 5 6 7 [START FIGURE 4.19 HERE] 8 9 Figure 4.19: Mid- and long-term change of annual mean surface temperature. Displayed are projected spatial 10 patterns of multi-model mean change in annual mean near-surface air temperature (°C) in 2041–2060 and 11 2081–2100 relative to 1995–2014 for (top) SSP1-2.6 and (bottom) SSP3-7.0. The number of models used 12 is indicated in the top right of the maps. No overlay indicates regions where the change is robust and 13 likely emerges from internal variability, that is, where at least 66% of the models show a change greater 14 than the internal-variability threshold (see Section 4.2.6) and at least 80% of the models agree on the sign 15 of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer 16 than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate 17 areas of conflicting signals where at least 66% of the models show change greater than the internal- 18 variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data 19 sources and processing are available in the chapter data table (Table 4.SM.1). 20 21 [END FIGURE 4.19 HERE] 22 23 24 The multi-model mean temperature change pattern (Figure 4.19) shows some robust key characteristics that 25 are independent of the time horizon and scenario, such as a land-ocean warming contrast, amplified warming 26 over the Arctic region, assessed below, or the comparatively small warming or even cooling in the North 27 Atlantic subpolar gyre (see Section 9.2.1.1). Changes in aerosol concentrations and land use and land 28 management can furthermore have a direct imprint on the regional warming pattern (Bright et al., 2017; 29 Kasoar et al., 2018). Note that the global average of the pattern shown in Figure 4.19 corresponds to CMIP6 30 multi-model mean GSAT warming (see Section 4.3.1) and is thus somewhat warmer than the warming 31 pattern consistent with the central estimate of the GSAT range assessed in Section 4.3.4. Since the regional 32 mean warming scales well with global warming levels independent of the emission scenario (see Section 33 4.2.4), the key characteristics of the spatial pattern assessed here are largely independent of the difference 34 between CMIP6 multi-model global mean and assessed global GSAT change. 35 36 Land–Ocean Warming Contrast 37 It is virtually certain that future average warming will be higher over land than over the ocean. SRCCL, 38 Section 2.2.1 (Jia et al., 2019b) assessed that it is certain that land temperatures have increased more than 39 global mean temperatures since the pre-industrial period. This so-called land-ocean warming contrast is a 40 striking feature of observed trends (Lambert and Chiang, 2007; Byrne and O’Gorman, 2018) and projected 41 changes in surface-air temperature (Sutton et al., 2007; Joshi and Gregory, 2008; Dong et al., 2009; Lambert 42 et al., 2011; Drost et al., 2012; Bayr and Dommenget, 2013; Byrne and O’Gorman, 2013b; Izumi et al., 43 2013; Joshi et al., 2013). Between 1979 and 2016, average temperature over land increased by 42% more 44 than over the ocean (Byrne and O’Gorman, 2018). A similar warming contrast is found in CMIP5 45 projections though with large differences across models and latitudes (Sutton et al., 2007; Drost et al., 2012; 46 Byrne and O’Gorman, 2013b; Joshi et al., 2013), which is also consistent with paleoclimate evidence (Izumi 47 et al., 2013; Schmidt et al., 2014). The ratio of land-to-ocean warming is greater than one for almost all 48 regions (high confidence) and is larger for dry subtropical continents (about 1.5) than for moist regions in the 49 tropics and mid-latitudes (about 1.2) (Byrne and O’Gorman, 2013a). 50 51 Since AR5, a robust physical understanding of the warming contrast been developed. A simple theory based 52 on atmospheric dynamics and moisture transport shows that surface-air temperature and relative humidity 53 over land are strongly coupled, and demonstrates that the warming contrast occurs because air over land is 54 drier than over the ocean (Joshi et al., 2008; Byrne and O’Gorman, 2013a, 2013b, 2018). The warming 55 contrast causes land relative humidity to decrease (Byrne and O’Gorman, 2016, 2018; Chadwick et al., 2016) 56 and this feeds back on and strengthens the warming contrast. Differences in land relative humidity responses Do Not Cite, Quote or Distribute 4-54 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 across models are the primary cause of uncertainty in the land-ocean warming contrast (Byrne and 2 O’Gorman, 2013b). These land relative humidity changes are ultimately controlled by moisture transport 3 between the land and ocean boundary layers (Byrne and O’Gorman, 2016; Chadwick et al., 2016) and are 4 also sensitive to characteristics of land surfaces that are challenging to model, including stomatal 5 conductance and soil moisture (Berg et al., 2016; Zarakas et al., 2020). 6 7 Polar Amplification 8 It is very likely that under all SSPs the warming in the Arctic will be more pronounced than in the global 9 average over the 21st century. Since AR5 the understanding of the physical mechanisms driving polar 10 amplification has improved. 11 12 The Arctic surface is projected to warm more than the global average over the 21st century, with annual- 13 average Arctic warming of about 3°C (SSP1-2.6), 10°C (SSP3-7.0) and 12°C in (SSP5-8.5) by 2081–2100 14 relative to 1995–2014 (Figure 4.19). This phenomenon, known as polar or Arctic amplification, is a 15 ubiquitous feature of the response to GHG forcing simulated by climate models (Manabe and Wetherald, 16 1975; Manabe and Stouffer, 1980; Manabe and Wetherald, 1980; Robock, 1983; Hansen et al., 1984; 17 Manabe et al., 1991; Holland and Bitz, 2003; Winton, 2006; Pithan and Mauritsen, 2014) and has been 18 observed over recent decades concurrent with Arctic sea-ice loss (Serreze and Barry, 2011) (Chapter 2 19 Section 2.3.2.1). Based on robust scientific understanding and agreement across multiple lines of evidence 20 (Chapter 7 Section 7.4.4.1), there is high confidence that the rate of Arctic surface warming will continue to 21 exceed the global average over the 21st century. 22 23 A variety of mechanisms contribute to Arctic amplification (see Chapter 7 Section 7.4.4.1.1). While surface- 24 albedo feedbacks associated with the loss of sea ice and snow have long been known to play important roles 25 (Arrhenius, 1896; Manabe and Stouffer, 1980; Robock, 1983; Hall, 2004), it is now recognized that 26 temperature (lapse-rate and Planck) feedbacks also contribute to Arctic amplification through a less efficient 27 longwave radiative damping to space with warming at high latitudes (Winton, 2006; Pithan and Mauritsen, 28 2014; Goosse et al., 2018; Stuecker et al., 2018). Increases in poleward atmospheric latent heat transport and 29 oceanic heat transport also contribute to Arctic warming (Holland and Bitz, 2003; Bitz et al., 2006; Lee et 30 al., 2011, 2017; Alexeev and Jackson, 2013; Marshall et al., 2014, 2015; Woods and Caballero, 2016; Singh 31 et al., 2017; Nummelin et al., 2017; Oldenburg et al., 2018; Merlis and Henry, 2018; Armour et al., 32 2019)(Beer et al., 2020). Projected reduction in the strength of the AMOC over the 21st century is expected 33 to reduce Arctic warming, but even a strong AMOC reduction would not eliminate Arctic amplification 34 entirely (Liu et al., 2017, 2018d; Wen et al., 2018) (medium confidence). 35 36 There remains substantial uncertainty in the magnitude of projected Arctic amplification (Smith et al., 2020), 37 with the Arctic warming ranging from two to four times the global average in models (Holland and Bitz, 38 2003; Nummelin et al., 2017). This uncertainty primarily stems from different representations of polar 39 surface-albedo, lapse-rate, and cloud feedbacks, and from different projected poleward energy transport 40 changes (Holland and Bitz, 2003; Crook et al., 2011; Mahlstein and Knutti, 2011; Pithan and Mauritsen, 41 2014; Bonan et al., 2018). The magnitude of Arctic amplification may also depend on the mix of radiative 42 forcing agents (Najafi et al., 2015; Sand et al., 2016; Stjern et al., 2019) such as the contribution of ozone 43 depleting substances (Polvani et al., 2020). Tropospheric aerosol emissions tend to reduce simulated Arctic 44 warming over the middle of the 20th century (Gagné et al., 2017a) and consequently aerosol emission 45 reductions in observations and SSP scenarios enhance simulated Arctic warming over recent and future 46 decades (Gagné et al., 2015; Acosta Navarro et al., 2016; Wobus et al., 2016; Wang et al., 2018) (also see 47 Chapter 6 Section 6.4.3). 48 49 Climate models project weakly polar amplified warming in the SH under transient warming (Figure 4.19). 50 Model simulations (Hall, 2004; Danabasoglu and Gent, 2009; Li et al., 2013) and paleoclimate proxies 51 indicate polar amplification in both hemispheres near equilibrium, but generally with less warming in the 52 Antarctic than the Arctic (Chapter 7, Section 7.4.4.1.2). The primary driver of delayed warming of the 53 southern high latitudes is the upwelling in the Southern Ocean and associated ocean heat uptake that is then 54 transported away from Antarctica by northward flowing surface waters (Froelicher et al., 2015; Marshall et 55 al., 2015; Armour et al., 2016; Liu et al., 2018c), although asymmetries in feedbacks between the poles also Do Not Cite, Quote or Distribute 4-55 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 play a role (Chapter 7, Section 7.4.4.1.1). Changes in westerly surface winds over the Southern Ocean have 2 the potential to affect the rate of sea-surface warming, but there is currently low confidence in even the sign 3 of the effect based on a diverse range of climate model responses to wind changes (Marshall et al., 2014; 4 Ferreira et al., 2015; Kostov et al., 2017; Seviour et al., 2019). A substantial increase in freshwater input to 5 the ocean from the Antarctic ice sheet could further slow the emergence of SH polar amplification by 6 cooling the Southern Ocean surface (Bronselaer et al., 2018; Golledge et al., 2019; Schloesser et al., 2019), 7 but this process is not represented in current climate models which lack dynamic ice sheets. Thus, while 8 there is high confidence that the SH high latitudes will warm by more than the tropics on centennial 9 timescales, there is low confidence that such a feature will emerge this century (Chapter 7, Section 7.4.4.1). 10 11 Seasonal Warming Patterns 12 The warming pattern shows distinct seasonal characteristics. The majority of models show a stronger 13 hemispheric winter than summer warming over land poleward of about 55°N and 55°S (Figure 4.20) and 14 thereby a reduced amplitude of the temperature cycle (Dwyer et al., 2012; Donohoe and Battisti, 2013). On 15 the other hand, over most of the subtropics and mid-latitudinal land regions except for parts of Asia, models 16 project stronger warming in hemispheric summer than winter (Donohoe and Battisti, 2013; Santer et al., 17 2018), leading to an amplification of the seasonal cycle. This phenomenon has been studied particularly in 18 the case of the amplified summer warming over the Mediterranean region (Seager et al., 2014a; Kröner et al., 19 2017; Brogli et al., 2019). 20 21 22 [START FIGURE 4.20 HERE] 23 24 Figure 4.20: Difference of surface temperature change between JJA and DJF. Displayed are spatial patterns of 25 multi-model mean difference in projected warming in JJA minus warming in DJF in 2081–2100 relative 26 to 1995–2014 for (left) SSP1-2.6 and (right) SSP3-7.0. Diagonal lines mark areas where fewer than 80% 27 of the models agree on the sign of change, and no overlay where at least 80% of the models agree. 28 Further details on data sources and processing are available in the chapter data table (Table 4.SM.1). 29 30 [END FIGURE 4.20 HERE] 31 32 33 Changes in Temperature Variability 34 It has long been recognized that along with mean temperatures also variance and skewness of the 35 temperature distribution may be changing (Gregory and Mitchell, 1995; Mearns et al., 1997). By amplifying 36 or dampening changes in the tail of temperature distribution such changes are potentially highly relevant to 37 extremes (Chapter11, Section 11.3.1) and pose a serious challenge to adaptation measures. Changes in 38 temperature variability can occur from diurnal to multi-decadal timescales and from the local to the global 39 scale with potentially even opposing signals in different seasons and at the different spatial scales 40 41 Changes in GSAT variability are poorly understood. Based on model experiments it has been suggested that 42 unforced variability of GSAT tends to decrease in a warmer world as a result of reduced albedo variability in 43 high latitudes resulting from melting snow and sea ice (Huntingford et al., 2013; Brown et al., 2017), but 44 confidence remains low and an observed change has not been detected. An assessment of changes in global 45 temperature variability is inherently challenging due to the interplay of unforced internal variability and 46 forced changes. 47 48 Changes in tropical temperature variability may arise from changes in the amplitude of ENSO (see Section 49 4.5.3.2). Over the extratropics, several studies have identified robust large-scale patterns of changes in 50 variability of annual and particularly seasonal mean temperature, including (a) a reduction in mid- to high- 51 latitude winter temperature variability and (b) an increase in summer temperature variability over land in the 52 tropics and subtropics (Huntingford et al., 2013; Holmes et al., 2016)(Figure 4.21). The multi-ensemble 53 average across seven single-model initial-condition large ensembles projects a consistent reduction in year- 54 to-year DJF variability around about 50-70°N and JJA variability around 55°–70°S along the edge of the sea 55 ice- and snow-covered region (Figure 4.21). There is growing evidence that year-to-year and day-to-day Do Not Cite, Quote or Distribute 4-56 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 temperature variability decreases in winter over northern mid- to high-latitudes (Fischer et al., 2011; De 2 Vries et al., 2012; Screen, 2014; Schneider et al., 2015; Holmes et al., 2016; Borodina et al., 2017; Tamarin- 3 Brodsky et al., 2020) which implies that the lowest temperatures rise more than the respective climatological 4 mean temperatures (medium confidence). Over the Northern Hemisphere, reduced high-latitude temperature 5 variability is associated with disproportionally large warming in source region of cold-air advection due to 6 Arctic amplification and land-sea contrast (De Vries et al., 2012; Screen, 2014; Holmes et al., 2016). It has 7 further been argued that a reduction in snow and sea-ice coverage from partly to completely snow- and ice- 8 free ocean and land surface would substantially reduce cold-season temperature variability (Gregory and 9 Mitchell, 1995; Fischer et al., 2011; Borodina et al., 2017) and lead to a shortening of the cold season and 10 earlier onset of the warm season (Cassou and Cattiaux, 2016). Mid-latitudinal winter temperature variability 11 is further affected by a complex interplay of a multitude of processes including potential changes in 12 atmospheric circulation, but there is low confidence in the dominant contribution of Arctic warming 13 compared to other drivers (see Cross-Chapter Box 10.1). 14 15 16 [START FIGURE 4.21 HERE] 17 18 Figure 4.21: Percentage change in interannual variability of (top) JJA and (bottom) DJF mean temperature 19 averaged across seven large initial condition ensembles. Average changes across seven single-model 20 initial-condition large ensembles are shown for RCP8.5 in 2081–2100 (and where not available for 2080– 21 2099) relative to 1995–2014. Standard deviations are calculated across all members of the large 22 ensembles for every given year to avoid inflation due to the underlying trend and then averaged across the 23 period. Changes are averaged across the ensembles MPI-GE (100 members, (Maher et al., 2019a)), 24 CanESM2, 50 members (Kirchmeier-Young et al., 2017)), NCAR-CESM (30 members, (Kay et al., 25 2015)), GFDL-CM3, 20 members, (Rodgers et al., 2015)), GFDL-ESM2M (30 members, (Sun et al., 26 2018)), CSIRO-Mk3-6-0 (30 members, (Jeffrey et al., 2013)), EC-EARTH (16 members, (Hazeleger et 27 al., 2010)), see (Deser et al., 2020). Diagonal lines indicate areas with low model agreement where fewer 28 than 80% of the models agree on the sign of the change, and no overlay areas with high model agreement 29 where at least 80% of the models agree on the sign of the change. Further details on data sources and 30 processing are available in the chapter data table (Table 4.SM.1). 31 32 [END FIGURE 4.21 HERE] 33 34 35 In JJA, the multi-model average projects an increase in year-to-year JJA variability over Central Europe and 36 North America (Figure 4.21). In particular an increase in daily to interannual summer temperature variability 37 has been projected over central Europe as a result of larger year-to-year variability in soil moisture 38 conditions varying between a wet and dry regime and leading to enhanced land-atmosphere interaction 39 (Seneviratne et al., 2006; Fischer et al., 2012; Holmes et al., 2016). Furthermore, the amplified warming in 40 the source regions of warm-air advection due to land-ocean warming contrast and amplified Mediterranean 41 warming (Seager et al., 2014a; Brogli et al., 2019), may lead to disproportionally strong warming of the 42 hottest days and summers and thereby increased variability. Enhanced temperature variability is further 43 projected over some land regions in the subtropics and tropics (Bathiany et al., 2018). 44 45 In summary, there is medium confidence that continued warming will regionally lead to increased and 46 decreased year-to-year temperature variability in the extratropics and there is medium confidence that year- 47 to-year temperature variability will decrease over parts of the mid- to high- latitudes of the winter 48 hemisphere. 49 50 51 4.5.1.2 Annual Mean Atmospheric Temperature 52 53 Section 12.4.3.2 of the AR5 assessed that there is high confidence in the overall pattern of projected end of 54 21st century tropospheric temperature change and that it is very likely that some of the largest warming will 55 occur in the northern high latitudes. They further assessed that proportionately larger warming is likely to 56 occur in the tropical upper troposphere than at the tropical surface, but with medium confidence owing to the Do Not Cite, Quote or Distribute 4-57 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 relatively large observational uncertainties and contradictory analyses regarding model accuracy in 2 simulating tropical upper tropospheric temperature trends. 3 4 5 [START FIGURE 4.22 HERE] 6 7 Figure 4.22: Long-term change of annual and zonal mean atmospheric temperature. Displayed are multi-model 8 mean change in annual and zonal mean atmospheric temperature (°C) in 2081–2100 relative to 1995– 9 2014 for (left) SSP1-2.6 and (right) SSP5-8.5. The number of models used is indicated in the top right of 10 the maps. Diagonal lines indicate regions where less than 80% of the models agree on the sign of the 11 change and no overlay where 80% or more of the models agree on the sign of the change. Further details 12 on data sources and processing are available in the chapter data table (Table 4.SM.1). 13 14 [END FIGURE 4.22 HERE] 15 16 17 CMIP6 projections show warming throughout the troposphere by the end of this century and a mix of 18 warming and cooling in the stratosphere depending on the emission scenario (Figure 4.22). The patterns of 19 tropospheric temperature change are highly consistent with those derived from earlier generations of climate 20 models as assessed in AR5, AR4 and TAR. In SSP1-2.6, the multi-model mean warming remains below 3°C 21 everywhere in the troposphere except near the surface in the Arctic; this is similar to the findings in AR5 22 based on CMIP5 models for RCP2.6. In SSP3-7.0, the zonal mean tropospheric warming is also largest in 23 the tropical upper troposphere, reaching more than 5 °C, and near the surface in the Arctic where warming 24 exceeds 8°C (Figure 4.22). It is likely that the warmer projected GSAT in the unconstrained CMIP6 model 25 ensemble contributes to larger warming in the tropical upper troposphere and in the Arctic lower 26 troposphere. This assessment is based on the understanding of polar amplification assessed in Chapter 7, 27 Section 7.4.4.1, and at low latitudes is based on the understanding of moist convective processes as well as 28 the relationship between CMIP5- and CMIP6-simulated surface temperatures and tropical upper tropospheric 29 warming over the historical period (Section 3.3.1.2). 30 31 Projected stratospheric temperature trends are determined by a balance between the major radiative drivers 32 from ozone recovery, rising CO2 and other greenhouse gases (including stratospheric water vapour) 33 (Maycock, 2016), as well as future changes in the Brewer Dobson circulation, which can alter the latitudinal 34 pattern of stratospheric temperature trends (Fu et al., 2015, 2019). In the lower stratosphere, the CMIP6 35 models project a weak cooling in the inner tropics in SSP1-2.6 and a warming at other latitudes (Figure 36 4.22). There is enhanced lower stratospheric warming over the Antarctic pole owing to the effects of ozone 37 hole recovery on polar temperatures (Maycock, 2016; Solomon et al., 2017). The projected strengthening of 38 the Brewer Dobson circulation in the future (Hardiman et al., 2014) also affects stratospheric temperature 39 trends, with adiabatic cooling at low latitudes and warming in middle and high latitudes (Fu et al., 2015, 40 2019). In SSP3-7.0, there is widespread cooling across much of the stratosphere, as expected from the higher 41 GHG emissions, with a smaller warming in the Antarctic lower stratosphere. Owing to the importance of 42 ozone recovery for the radiative balance of the stratosphere, future global and local stratospheric temperature 43 trends do not scale with projected GSAT change. 44 45 In summary, new results since AR5 do not generally alter the understanding of projected zonal mean 46 atmospheric temperature changes. There is high confidence in the overall pattern of projected tropospheric 47 temperature changes given its robustness across many generations of climate models. It is further very likely 48 that projected long-term tropospheric warming will be larger than the global mean in the Arctic lower 49 troposphere. It is likely that tropical upper tropospheric warming will be larger than at the tropical surface, 50 however with an uncertain magnitude owing to the potentially large role of natural internal variability and 51 differences across models in the simulated free tropospheric temperature response to a given forcing scenario 52 (Section 3.3.1.2). It is very likely that global mean stratospheric cooling will be larger by the end of the 21st 53 century in a pathway with higher atmospheric CO2 concentrations. 54 55 56 4.5.1.3 Near-Surface Relative Humidity Do Not Cite, Quote or Distribute 4-58 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 AR5 contrasted future changes in near-surface relative humidity (RH) over land and ocean, concluding with 3 medium confidence that reductions in near-surface RH over many land areas are likely. The decrease in near- 4 surface RH over most land areas is associated with the larger warming rates over land than over the ocean 5 and is termed the last-saturation-temperature constraint, as explained in AR5. 6 7 Since AR5, significant effort has been devoted to understanding the mechanisms for the decrease in near- 8 surface land RH under global warming, and the relevance of RH changes for the land–sea warming contrast 9 and the water cycle. For the near-surface RH decrease over land, both the moisture transport from the ocean 10 and land-atmosphere feedback processes contribute. For changes in specific humidity over land, the moisture 11 transport from the ocean is dominant while the role of evapotranspiration is secondary (Byrne and 12 O’Gorman, 2016; Chadwick et al., 2016). Nevertheless, the changes in near-surface land RH are also 13 strongly influenced by evapotranspiration, which is suppressed by the drying of soils and plant responses to 14 increasing CO2 related to stomatal closure under climate change (Byrne and O’Gorman, 2015; Berg et al., 15 2016; Chadwick et al., 2016; Swann et al., 2016; Lemordant et al., 2018). The combination of oceanic and 16 continental influences can explain the spatially diverse trends in the near-surface RH over land in the 17 observations for the recent decades, with a generally dominant negative trend at the global scale (Vicente- 18 Serrano et al., 2018). There is a strong feedback between the near-surface land RH decrease and land-ocean 19 warming contrast under future warming projections (see Section 4.5.1.1). 20 21 Changes in land RH can modulate the response of the water cycle to global warming (Chadwick et al., 2013; 22 Byrne and O’Gorman, 2015). Most CMIP5 models project higher precipitation associated with higher near- 23 surface RH and temperature under climate change (Lambert et al., 2017). Over land, the spatial gradients of 24 fractional changes in near-surface RH contribute to a drying tendency in precipitation minus 25 evapotranspiration with warming, which partly explains why the ‘wet-gets-wetter, dry-gets-drier’ principle 26 does not hold over land (Byrne and O’Gorman, 2015). Terrestrial aridity is projected to increase over land, 27 as manifested by a decrease in the ratio of precipitation to potential evapotranspiration, in which the decrease 28 in near-surface land RH has a contribution of about 35% in CMIP5 models under doubled CO2 forcing (Fu 29 and Feng, 2014). The aridity can be further amplified by the feedbacks of projected drier soils on land 30 surface temperature, RH, and precipitation (Berg et al., 2016). 31 32 The CMIP6 multi-model ensemble projects general decreases in near-surface relative humidity over a large 33 fraction of land areas, but moderate increases over the ocean (Figure 4.23). The projected changes depend on 34 emission scenario and season. Changes in near-surface RH under SSP1-2.6 are insignificant compared to 35 natural variability. Under SSP3-7.0, during boreal summer, significant decreases relative to natural 36 variability are projected in continental Europe and the Middle East, North America, South America and 37 South Africa. 38 39 In summary, there is medium confidence that continued warming will lead to decreased near-surface relative 40 humidity over a large fraction of land areas, but moderate increases over the ocean. There is high confidence 41 that near-surface relative humidity will decrease over parts of the tropical and subtropical latitudes over land. 42 43 44 [START FIGURE 4.23 HERE] 45 46 Figure 4.23: Long-term changes in seasonal mean relative humidity. Displayed are projected spatial patterns of 47 multi-model mean change (%) in seasonal (top) DJF and (bottom) JJA mean near-surface relative 48 humidity in 2081–2100 relative to 1995–2014, for (left) SSP1-2.6 and (right) SSP3-7.0. The number of 49 models used is indicated in the top right of the maps. No overlay indicates regions where the change is 50 robust and likely emerges from internal variability, that is, where at least 66% of the models show a 51 change greater than the internal-variability threshold (see Section 4.2.6) and at least 80% of the models 52 agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant 53 change, where fewer than 66% of the models show change greater than the internal-variability threshold. 54 Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater 55 than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. 56 Further details on data sources and processing are available in the chapter data table (Table 4.SM.1). Do Not Cite, Quote or Distribute 4-59 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 [END FIGURE 4.23 HERE] 3 4 5 4.5.1.4 Precipitation 6 7 AR5 assessed that changes in mean precipitation in a warmer world will exhibit substantial spatial variation 8 and the contrast of mean precipitation between dry and wet regions and between dry and wet seasons will 9 increase over most of globe as temperature increase. The general pattern of change indicates that high 10 latitude land masses are likely to experience greater amounts of precipitation due to the increased specific 11 humidity of the warmer troposphere as well as increased transport of water vapour from the tropics by the 12 end of this century under the RCP8.5 scenario. Many mid-latitude and subtropical arid and semi-arid regions 13 will likely experience less precipitation and many moist mid-latitude regions will likely experience more 14 precipitation by the end of this century under the RCP8.5 scenario. 15 16 Since AR5, progress has been achieved in understanding changes in patterns and rates of precipitation with 17 GSAT rise. The projected precipitation changes can be decomposed into a part that is related to atmospheric 18 circulation referred to as dynamical component and a part related to water vapour changes, the 19 thermodynamic component. Based on process understanding and modelling (Fläschner et al., 2016; Samset 20 et al., 2016), global mean precipitation will very likely increase by 1–3% per °C of GSAT warming (see 21 Section 8.2.1). The increase in atmospheric water vapour is a robust change under global warming, the 22 sensitivity of global precipitation change to warming is smaller (2% °C-1) as compared to water vapour 23 change (7% °C-1) (Held and Soden, 2006a). Global energy balance places a strong constraint on the global 24 mean precipitation (Allen and Ingram, 2002; Pendergrass and Hartmann, 2014; Myhre et al., 2018; Siler et 25 al., 2019). Tropospheric radiative cooling constrains global precipitation (Pendergrass and Hartmann, 2014), 26 leading to a slow SST-dependent response and a forcing-dependent rapid adjustment. Rapid adjustments 27 account for large regional differences in hydrological sensitivity across multiple drivers (Samset et al., 2016; 28 Myhre et al., 2017). The rapid regional precipitation response to increased CO2 is robust across models, 29 implying that the uncertainty in long-term changes is mainly associated with the response to SST-mediated 30 feedbacks (Richardson et al., 2016). Precipitation response to fast adjustments and slow temperature-driven 31 responses are assessed in detail in Chapter 8 Section 8.2.1. 32 33 The thermodynamic response to global warming is associated with a wet-get-wetter mechanism, with 34 enhanced moisture flux leading to subtropical dry regions getting drier and tropical and mid-latitude wet 35 regions getting wetter (Held and Soden, 2006a; Chou et al., 2009). Recent studies suggest that the dry-get- 36 drier argument does not hold, especially over subtropical land regions (Greve et al., 2014; Feng and Zhang, 37 2015; Greve and Seneviratne, 2015). The discrepancy may be partly arising due to differences in model 38 climatologies and by change in the location of wet and dry regions (Polson and Hegerl, 2017). Over the 21st 39 century, significant rate of precipitation change is associated with a spatial stabilization and intensification of 40 moistening and drying patterns (Chavaillaz et al., 2016a). In the tropics, weakening of circulation leads to a 41 wet-gets-drier and dry-gets-wetter pattern (Chadwick et al., 2013). Climate model agreement for 42 precipitation change in the tropics is lower than for other regions (Knutti and Sedláček, 2013; McSweeney 43 and Jones, 2013). Sources of inter-model uncertainty in regional rainfall projections arise from circulation 44 changes (Kent et al., 2015; Chadwick, 2016) and spatial shifts in convection and convergence, associated 45 with SST pattern change and land–sea thermal contrast change (Kent et al., 2015; Chadwick et al., 2017) 46 with a secondary contribution from the response to direct CO2 forcing (Chadwick, 2016). Factors governing 47 changes in large-scale precipitation patterns are assessed in detail in Section 8.2.2 and Section 10.4.1. 48 49 Long-term multi-model mean change in seasonal precipitation (JJA and DJF) from CMIP6 models (Figure 50 4.24) shows substantial regional differences and seasonal contrast. Changes in seasonal precipitation under 51 SSP1-2.6 are small compared to internal variability. Consistent with the AR5, patterns of precipitation 52 change show very likely increase in the high latitudes especially during local winter and over tropical 53 oceanoceans under SSP3-7.0 (high confidence). CMIP6 projections show an increase in precipitation over 54 larger parts of the monsoon regions and decreases in many subtropical regions including the Mediterranean, 55 southern Africa and southwest Australia (medium confidence). The large-scale patterns of precipitation Do Not Cite, Quote or Distribute 4-60 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 shown in Figure 4.24 are consistent with the patterns presented in Section 8.4.1.3. Precipitation changes 2 exhibit strong seasonal characteristics (Box 8.2), and, in many regions, the sign of the precipitation changes 3 varies with season. Precipitation variability is projected to increase over a majority of global land area, as 4 assessed in Chapter 8 Section 8.4.1.3.3, over a wide range of timescales in response to warming (Pendergrass 5 et al., 2017). 6 7 8 [START FIGURE 4.24 HERE] 9 10 Figure 4.24: Long-term change of seasonal mean precipitation. Displayed are projected spatial patterns of multi- 11 model mean change (%) in (top) DJF and (bottom) JJA mean precipitation in 2081–2100 relative to 12 1995–2014, for (left) SSP1-2.6 and (right) SSP3-7.0. The number of models used is indicated in the top 13 right of the maps. No map overlay indicates regions where the change is robust and likely emerges from 14 internal variability, that is, where at least 66% of the models show a change greater than the internal- 15 variability threshold (see Section 4.2.6) and at least 80% of the models agree on the sign of change. 16 Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of 17 the models show change greater than the internal-variability threshold. Crossed lines indicate areas of 18 conflicting signals where at least 66% of the models show change greater than the internal-variability 19 threshold but fewer than 80% of all models agree on the sign of change. Further details on data sources 20 and processing are available in the chapter data table (Table 4.SM.1). 21 22 [END FIGURE 4.24 HERE] 23 24 25 Most of the projected changes in precipitation exhibit a sharp contrast between land and ocean (see Sections 26 8.2.1 and 8.4.1). Temperature-driven intensification of land-mean precipitation during the 20th century has 27 been masked by fast precipitation responses to anthropogenic sulphate and volcanic forcing (Allen and 28 Ingram, 2002; Richardson et al., 2018b). Based on the Precipitation Driver and Response Model 29 Intercomparison Project (PDRMIP), land-mean precipitation is expected to increase more rapidly with the 30 projected decrease in sulphate forcing and continued warming, contributing to increase global mean 31 precipitation (Table 4.3) and will be clearly observable by the mid-21st century based on RCP4.5 and 32 RCP8.5 scenarios (Richardson et al., 2018b). 33 34 Consistent with the findings of AR5, a gradual increase in global mean precipitation is projected over the 35 21st century with an increase of approximately 2.9% (1.0–5.2%) under SSP1-2.6 and 4.7% (2.3–8.2%) under 36 SSP3-7.0 during 2081–2100 relative to 1995–2014. The corresponding increase in annual mean global land 37 precipitation is 3.3% (0–6.6%), in the SSP1-2.6 and 5.8% (0.5–9.6%) in the SSP3-7.0. (See also Table 4.3). 38 CMIP6 models show greater increases in precipitation over land than either globally or over the ocean (high 39 confidence). 40 41 Based on the assessment of CMIP6 models, we conclude that it is very likely that, in the long term, global 42 mean land and global mean ocean precipitation will increase with increasing GSAT. Annual mean and global 43 mean precipitation will very likely increase by 1–3% per °C GSAT warming. The patterns of precipitation 44 change will exhibit substantial regional differences and seasonal contrast as GSAT increases over the 21st 45 century (high confidence). Precipitation will very likely increase over high latitudes and the tropical ocean 46 and likely increase in large parts of the monsoon regions, but likely decrease over the subtropics, including 47 Mediterranean, southern Africa and southwest Australia, in response to GHG-induced warming. 48 49 50 4.5.1.5 Global Monsoon Precipitation and Circulation 51 52 AR5 assessed changes of the global monsoon in the context of long-term trends across the 21st century and 53 the change by 2081–2100. AR5 showed growing evidence of improved skill of climate models in 54 reproducing the climatological features of the global monsoon. Taken together with identified model 55 agreement on future changes, the global monsoon precipitation, aggregated over all regional monsoon 56 regions, is likely to strengthen in the 21st century with increases in its area and intensity, while the monsoon Do Not Cite, Quote or Distribute 4-61 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 circulation weakens. In all RCP scenarios, the global monsoon area is very likely to increase, and the global 2 monsoon precipitation intensity is likely to increase, resulting in a very likely increase in the global monsoon 3 total precipitation, by 2081–2100 (Kitoh et al., 2013). 4 5 Since AR5, there has been progress in understanding physical mechanisms for the projected changes in 6 global monsoon and quantifying the sources of uncertainty in projections. The increase in global monsoon 7 precipitation under warming is primarily attributed to the increase of moisture convergence, which comes 8 mainly from the thermodynamic effect due to increasing atmospheric moisture but is partly offset by reduced 9 convergence (Zhang et al., 2019b; Chen et al., 2020). The dynamic effect, such as monsoon circulation 10 changes, dominates regional differences in the projected monsoon precipitation changes (Chen et al., 2020). 11 Specifically, NH monsoon precipitation will increase more strongly than its SH counterpart, due to an 12 increase in hemispheric temperature difference between the NH and SH, enhancement of the Hadley 13 circulation, and atmospheric moistening, countered by stabilization of the troposphere (Lee and Wang, 14 2014). The seasonality of global monsoon rainfall is projected to enhance in response to warming, featuring 15 a greater wet-dry season contrast (Lee and Wang 2014; Zhang et al. 2019). In addition, the interannual 16 variability of global monsoon rainfall is projected to intensify mainly over land, with a strengthened 17 relationship between global monsoon and ENSO (Hsu et al., 2013; Wang et al., 2020, 2021). 18 19 For the uncertainty in mean monsoon precipitation projections, the model uncertainty is the dominant 20 contributor throughout the century and explains more than 70% of the inter-model variance during near term, 21 mid-term, and long term. The contribution of internal variability is only important at the beginning in early 22 decades, while scenario uncertainty becomes important at the end of the 21st century. The sources of 23 uncertainty for the mean and extreme monsoon precipitation mainly differ in the long-term projection, when 24 the contribution of scenario uncertainty is comparable to the model uncertainty for extreme precipitation 25 (Zhou et al., 2020). Although the magnitude of internal variability differs between CMIP5 models and 26 single-model initial-condition large ensembles, the impact is only evident in the beginning decades. For the 27 mid- and long term, the magnitude difference does not alter that model uncertainty is the dominant source of 28 uncertainty in the projections of global land monsoon precipitation (Zhou et al., 2020). 29 30 Based on the projections of changes in precipitation from CMIP6 under the four SSPs, the global monsoon 31 precipitation is likely to strengthen in the 21st century with increases in its intensity, while NH summer 32 monsoon circulation weakens (Figure 4.14). Global land monsoon precipitation will likely increase by 1.3– 33 2.4 % per °C GSAT warming among the four scenarios considered here. In the long term, the multi-model 34 mean change (5–95% range of the available 41 projections) of global land monsoon precipitation index is 35 2.9% (–0.8–7.8%), 3.7% (–2.5–8.6%), 3.77% (–3.2–8.1%), and 5.7% (–2.8–12.3%) under SSP1–2.6, SSP2– 36 4.5, SSP3–7.0, and SSP5–8.5, respectively. This enhancement is caused by thermodynamic responses due to 37 increased moisture, which is partly offset by dynamic responses due to a weakened circulation (Chen et al., 38 2020). The patterns of monsoon rainfall changes in the mid- to long-term include a North–South asymmetry 39 characterized by greater increase in the NH than the SH, and an East–West asymmetry characterized by 40 enhanced Asian-African monsoons and weakened North American monsoon (medium confidence) (Lee and 41 Wang, 2014; Mohtadi et al., 2016; Pascale et al., 2017; Wang et al., 2021). 42 43 Based on the assessment of CMIP6 models, we conclude that it is likely that, in the mid- to long term, the 44 global land monsoon precipitation will increase with GSAT rise despite a weakened monsoon circulation. 45 The global land monsoon precipitation will likely increase by 1.3–2.4 % per °C GSAT warming among the 46 four scenarios. Monsoon precipitation responses depend on region and emission scenario (high confidence). 47 48 49 4.5.1.6 Sea Level Pressure, Large-scale Atmospheric Circulation, Storm Tracks and Blocking 50 51 This subsection provides a global overview of long-term changes in atmospheric dynamical features that is 52 complementary to the regional assessment of links to the hydrological cycle in Chapter 8, Section 8.4.2, and 53 assessment of the connections to extreme events in Chapter 11, Section 11.7.2. 54 55 Sea level pressure Do Not Cite, Quote or Distribute 4-62 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 AR5 assessed that mean sea level pressure is projected to decrease in high latitudes and to increase in mid- 2 latitudes. Such a pattern is associated with a poleward shift in the storm track and an increase in the annular 3 mode index. This broad pattern is also found in CMIP6 models (Figure 4.25). Under SSP1-2.6, the pattern in 4 sea level pressure change resembles that for SSP3-7.0, but the amplitudes are small compared to internal 5 variability in 20-year means (Figure 4.25). One exception is found in the SH mid-latitudes, where pressure 6 robustly increases in SSP3-7.0 in both austral summer and winter, but shows no robust change in SSP1-2.6. 7 This is likely attributable to the larger GHG forcing in SSP3-7.0 compared to SSP1-2.6, which contributes to 8 a poleward shift of the SH mid-latitude circulation and becomes relatively more important than the effect of 9 ozone recovery which drives an equatorward shift in the circulation (Barnes and Polvani, 2013; Barnes et al., 10 2014; Bracegirdle et al., 2020b) (see Section 4.5.3.1 on the Southern Annular Mode). The poleward shift in 11 SH mid-latitude circulation in SSP3-7.0 likely contributes to the wetting trend at high southern latitudes 12 (Figure 4.25). 13 14 As was found in AR5, several regional sea level pressure features stand out from the zonal-mean change. Sea 15 level pressure markedly decreases in northeastern North America and northeastern Asia in boreal winter. In 16 boreal summer, sea level pressure robustly decreases in the Mediterranean and the Middle-East, a decrease 17 that has been linked to a large-scale heat low forced by the amplified warming of the region (Haarsma et al., 18 2009). It is likely that sea level pressure will increase across the southwestern North America and Central 19 America in boreal summer under SSP3-7.0 due to an intensification of the eastern North Pacific subtropical 20 summer high (Li et al., 2012) and a weakening of the North American monsoon (Wang et al., 2020)(Pascale 21 et al., 2017)(see Section 4.5.1.5). These changes in circulation are connected to drying across the eastern 22 subtropical Pacific and Central America regions (Figure 4.24:). 23 24 25 [START FIGURE 4.25 HERE] 26 27 Figure 4.25: Long-term change of seasonal mean sea level pressure. Displayed are projected spatial patterns of 28 multi-model mean change in (top) DJF and (bottom) JJA mean sea level pressure (hPa) in 2081–2100 29 relative to 1995–2014, for (left) SSP1-2.6 and (right) SSP3-7.0. The number of models used is indicated 30 in the top right of the maps. No overlay indicates regions where the change is robust and likely emerges 31 from internal variability, that is, where at least 66% of the models show a change greater than the 32 internal-variability threshold (see Section 4.2.6) and at least 80% of the models agree on the sign of 33 change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 34 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate 35 areas of conflicting signals where at least 66% of the models show change greater than the internal- 36 variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data 37 sources and processing are available in the chapter data table (Table 4.SM.1). 38 39 40 [END FIGURE 4.25 HERE] 41 42 43 Zonal wind and westerly jets 44 45 Storm tracks and mid-latitude westerly jets are dynamically related aspects of mid-latitude circulation. AR5 46 assessed that a poleward shift of the SH westerlies and storm track is likely by the end of the 21st century 47 under RCP8.5 (medium confidence). In contrast, low confidence was assessed for the storm-track response in 48 the NH. 49 50 Under both SSP1-2.6 and SSP3-7.0 there is a strengthening and lifting of the subtropical jets in both 51 hemispheres (Figure 4.26), consistent with the response to large-scale tropospheric warming found in earlier 52 generations of climate models (Collins et al., 2013). In the SH, GHG emissions tend to force a poleward shift 53 of the jet, but this is opposed, particularly in austral summer, by the stratospheric ozone hole recovery 54 (Barnes and Polvani, 2013; Barnes et al., 2014; Bracegirdle et al., 2020b). Consistent with sea level pressure 55 changes, CMIP6 models project a strengthening and poleward shift of the SH jet in austral summer and 56 winter under SSP3-7.0, but smaller and non-robust changes in SH mid-latitude zonal winds under SSP1-2.6 Do Not Cite, Quote or Distribute 4-63 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 (Figure 4.26, see also Section 4.5.3.1). CMIP6 models show an improved simulation of the SH jet stream 2 latitude (Bracegirdle et al., 2020a; Curtis et al., 2020). This has been linked to a reduction in the projected 3 poleward shift of the SH jet in austral summer compared to the CMIP5 models (Curtis et al., 2020; Goyal et 4 al., 2021), although differences in the pattern of SST response may also play a role (Wood et al., 2020). In 5 the NH extratropics, the changes in lower tropospheric zonal mean zonal winds by the end of the century are 6 generally smaller than in the SH. In boreal winter, there is a weak poleward shift of the NH zonal mean 7 westerly jet maximum in SSP3-7.0. 8 9 CMIP5 and CMIP6 models show a strong seasonal and regional dependence in the response to climate 10 change of NH westerlies (Barnes and Polvani, 2013; Grise and Polvani, 2014b; Simpson et al., 2014; Zappa 11 et al., 2015; Harvey et al., 2020; Oudar et al., 2020). CMIP5 projections indicate a poleward shift of the 12 westerlies in the North Atlantic in boreal summer, while the North Pacific jet weakens in this season 13 (Simpson et al., 2014; Davini and D’Andrea, 2020; Harvey et al., 2020). There is a poleward shift in the 14 westerlies in both the North Pacific and North Atlantic in Autumn (Barnes and Polvani, 2013; Simpson et al., 15 2014). However, the shift of the westerlies is more uncertain in the other seasons particularly in the North 16 Atlantic in winter (Simpson et al., 2014; Zappa and Shepherd, 2017). Here, the circulation response is not 17 well described as a simple shift, since the North Atlantic jet tends to be squeezed on both its equatorward and 18 poleward flanks, together with an eastward extension into Europe (Li et al., 2018; Peings et al., 2018; 19 Simpson et al., 2019b; Harvey et al., 2020; Oudar et al., 2020). Simulations indicate that most of the changes 20 in winter storminess over the Euro-Atlantic region will occur only after exceeding the 1.5°C warming level 21 (Barcikowska et al., 2018). 22 23 24 [START FIGURE 4.26 HERE] 25 26 Figure 4.26: Long-term change of zonal mean zonal wind. Dispayed are multi-model mean change in (left) boreal 27 winter (DJF) and (right) austral winter (JJA) zonal mean zonal wind (m s-1) in 2081–2100 for (top) SSP1- 28 2.6 and (right) SSP3-7.0 relative to 1995–2014. The 1995–2014 climatology is shown in contours with 29 spacing 10 m s-1. Crossed lines indicate regions where less than 80% of the models agree on the sign of 30 the change and no overlay where at least 80% of the models agree on the sign of the change. Further 31 details on data sources and processing are available in the chapter data table (Table 4.SM.1). 32 33 [END FIGURE 4.26 HERE] 34 35 36 Progress since AR5 has improved understanding of the climate change aspects that can drive these different, 37 and potentially opposite, responses in the mid-latitude jets and storm tracks. A poleward shift of the jets and 38 storm tracks is expected in response to an increase in the atmospheric stratification and in the upper- 39 tropospheric equator-to-pole meridional temperature gradient, while it is opposed by the decrease in the 40 meridional temperature gradient in the lower troposphere associated with the polar amplification of global 41 warming (Harvey et al., 2014; Shaw et al., 2016b). Recent analyses have identified additional climate aspects 42 that can drive mid-latitude jet changes, including patterns in sea surface warming (Mizuta et al., 2014; 43 Langenbrunner et al., 2015; Ceppi et al., 2018; Wood et al., 2020), land-sea warming contrast (Shaw and 44 Voigt, 2015), loss of sea ice (Zappa et al., 2018)(Deser et al., 2015)(Harvey et al., 2015)(Screen et al., 45 2018b), and changes in the strength of the stratospheric polar vortex (Simpson et al., 2018b)(Manzini et al., 46 2014)(Grise and Polvani, 2017)(Ceppi and Shepherd, 2019a). From an energetics perspective, the 47 uncertainty in the response of the jet streams depends on the response of clouds, their non-spatially uniform 48 radiative feedbacks shaping the meridional profile of warming (Ceppi et al., 2014; Voigt and Shaw, 2015, 49 2016; Ceppi and Hartmann, 2016; Ceppi and Shepherd, 2017; Lipat et al., 2018; Albern et al., 2019; Voigt et 50 al., 2019). Climate models seem to underestimate the forced component of the year-to-year variability in the 51 atmospheric circulation, particularly in the North Atlantic sector (Scaife and Smith, 2018), which suggests 52 some relevant dynamical processes may not be well represented. Whether and how this may affect long-term 53 projections is unknown. In conclusion, due to the influence from competing dynamical drivers and the 54 absence of observational evidence, there is medium confidence in a projected poleward shift of the NH 55 zonal-mean low-level westerlies in autumn and summer and low confidence in the other seasons. There is 56 also overall low confidence in projected regional changes in the NH low-level westerlies, particularly for the Do Not Cite, Quote or Distribute 4-64 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 North Atlantic basin in boreal winter. 2 3 The anthropogenic forced signal in extratropical atmospheric circulation may well be small compared to 4 internal variability (Deser et al., 2012b, 2014) and, as assessed in AR5, there is generally low agreement 5 across models in many aspects of regional atmospheric circulation change particularly in the NH (Shepherd, 6 2014). The latter means that, in some regions, a multi-model average perspective of atmospheric circulation 7 change represents a small residual after averaging over large intermodel spread. This is in strong contrast to 8 thermodynamic aspects of climate change, such as surface temperature change, for which model results are 9 generally highly consistent (see, e.g., Figure 4.19). Furthermore, models share systematic biases in some 10 aspects of extratropical atmospheric circulation such as mid-latitude jets, which can have complex 11 implications for understanding forced changes (Simpson and Polvani, 2016a). Given these issues, an 12 emerging field of research since AR5 has focused on the development of ‘storylines’ for regional 13 atmospheric circulation change (Shepherd, 2019). The storyline approach is grounded in the identification of 14 a set of physical predictors of atmospheric circulation change, such as those described above (Harvey et al., 15 2014; Manzini et al., 2014; Shepherd et al., 2018), which act together to determine a specific outcome in 16 theprojected atmospheric circulation change. The consequences of multi-model spread in the physical 17 predictors of atmospheric circulation change can be investigated, conditioned on a specified level of global 18 warming (Zappa and Shepherd, 2017; Zappa, 2019; Mindlin et al., 2020) (also see Chapter 1, Section 19 1.4.4.2). 20 21 Storm tracks 22 As stated in AR5, the number of extratropical cyclones (ETC) composing the storm tracks is projected to 23 weakly decline in future projections, but by no more than a few percent change. The reduction is mostly 24 located on the equatorward flank of the storm tracks, which is associated with the Hadley cell expansion and 25 a poleward shift in the mean genesis latitude of ETCs (Tamarin-Brodsky and Kaspi, 2017). Furthermore, the 26 poleward propagation of individual ETCs is expected to increase with warming (Graff and LaCasce, 2014; 27 Tamarin-Brodsky and Kaspi, 2017), thus contributing to a poleward shift in the mid-latitude transient-eddy 28 kinetic energy. The increased poleward propagation results from the strengthening of the upper tropospheric 29 jet and increased cyclone-associated precipitation (Tamarin-Brodsky and Kaspi, 2017), which are robust 30 aspects of climate change. 31 32 In the NH boreal winter, CMIP6 models show a northward shift of the ETC density in the North Pacific, a 33 tripolar pattern in the North Atlantic, and a weakening of the Mediterranean storm track (Figure 4.27a). 34 CMIP6 models show overall low agreement on changes in ETC density in the North Atlantic in boreal 35 winter (Figure 4.27a). A poleward shift of the storm track is evident in the SH (Figure 4.27b), particularly in 36 the Indian and Pacific Ocean sectors. CMIP6 models still feature long-standing biases in the representation 37 of storm tracks, such as a too zonal winter storm track into Europe, though different measures of storm track 38 activity indicate some improvements compared to the previous generations of models (Harvey et al., 2020; 39 Priestley et al., 2020) 40 41 Regarding the dynamical intensity of the storm tracks (see also Chapter 11, Section 11.7.2), the number of 42 ETCs associated with intense surface wind speeds and undergoing explosive pressure deepening are 43 projected to strongly decrease in the NH winter (Seiler and Zwiers, 2016; Chang, 2018). The weakening of 44 surface winds of ETCs in the NH is attributed to the reduced low-level baroclinicity from SST and sea ice 45 changes (Harvey et al., 2014; Seiler and Zwiers, 2016; Wang et al., 2017a). There are, however, regional 46 exceptions such as in the northern North Pacific, where explosive and intense ETCs are projected to increase 47 in association with the poleward shift of the jet and increased upper-level baroclinicity (Seiler and Zwiers, 48 2016). Eddy kinetic energy and intense cyclone activity is also projected to decrease in the NH summer in 49 association with a weakening of the jet (Lehmann et al., 2014; Chang et al., 2016). However, climate models 50 tend to have too weak explosive cyclones (Seiler and Zwiers, 2016; Priestley et al., 2020), though this bias 51 seems to be reduced in high-resolution simulations (Jiaxiang et al., 2020). Furthermore, models may not 52 fully capture the contribution of the future increase in mesoscale latent heating to cyclone intensification (Li 53 et al., 2014; Pfahl et al., 2015; Willison et al., 2015; Michaelis et al., 2017). In conclusion, there is only 54 medium confidence in the projected decrease in the frequency of intense NH ETCs. 55 Do Not Cite, Quote or Distribute 4-65 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 In contrast to the NH, the SH shows an increase in the frequency of intense ETCs in CMIP5 models (Chang, 2 2017), and there is high confidence that wind speeds associated with ETCs are expected to intensify in the 3 SH storm track for high emission scenarios. These changes in intensity are accompanied by an overall 4 southward shift of the SH winter storm track (Figure 4.27b) due to the poleward shift in the upper-level jet 5 and the increase in the meridional SST gradient linked to the slower warming of the Southern Ocean 6 (Grieger et al., 2014) . 7 8 Regardless of dynamical intensity changes, there is high confidence that the number of ETCs associated with 9 extreme precipitation is projected to increase with warming, due to the increase moisture-loading capacity of 10 the atmosphere (Yettella and Kay, 2017; Hawcroft et al., 2018) (see also Chapter 8, Section 8.4.2). 11 12 13 [START FIGURE 4.27 HERE] 14 15 Figure 4.27: Changes in extratropical storm track density. Displayed are projected spatial pattern of multi-model 16 mean change of extratropical storm track density in winter (NH DJF and SH JJA) in 2080–2100 for 17 SSP5-8.5 relative to 1979–2014 based on 13 CMIP6 models. Diagonal lines indicate regions where fewer 18 than 80% of the models agree on the sign of the change and no overlay where at least 80% of the models 19 agree on the sign of change. Units are number density per 5 degree spherical cap per month. Further 20 details on data sources and processing are available in the chapter data table (Table 4.SM.1). 21 22 [END FIGURE 4.27 HERE] 23 24 25 Atmospheric blocking 26 Blocking is associated with a class of quasi-stationary high-pressure weather systems in the middle and high 27 latitudes that disrupt the prevailing westerly flow. These events can persist for extended periods, such as a 28 week or longer, and can cause long-lived extreme weather conditions, from heat waves in summer to cold 29 spells in winter (see Chapter 11, Section 11.7.2 for a detailed discussion of these features and Chapter 3, 30 Section 3.3.3.3 for the assessment of blocking biases in models simulations). AR5 assessed with medium 31 confidence that the frequency of blocking would not increase under enhanced GHG concentrations, while 32 changes in blocking intensity and persistence remained uncertain. 33 34 CMIP5 projections suggest that the response of blocking frequency to climate change might be quite 35 complex (Dunn-Sigouin et al., 2013; Masato et al., 2013). An eastward shift of winter blocking activity in 36 the NH is indicated (Masato et al., 2013; Kitano and Yamada, 2016; Lee and Ahn, 2017; Matsueda and 37 Endo, 2017) while during boreal summer, blocking frequency tends to decrease in mid-latitudes (Matsueda 38 and Endo, 2017), with the exception of the eastern Europe–western Russia region (Masato et al., 2013). The 39 projected decrease of blocking in boreal summer partially contrasts with the observed increase in Greenland 40 blocking (Hanna et al., 2018; Davini and D’Andrea, 2020). However, as shown in Woollings et al. (2018), 41 the spatial distribution and the magnitude of the suggested changes are sensitive to the blocking detection 42 methods (Schwierz et al., 2004; Barriopedro et al., 2010; Davini et al., 2012). In the SH, blocking frequency 43 is projected to decrease in the Pacific sector during austral spring and summer. However, seasonal and 44 regional changes are not totally consistent across the models (Parsons et al., 2016), and, as assessed in 45 Section 3.3.3.3, model biases might affect their response. 46 47 To better understand the uncertainty in future blocking activity, a process-oriented approach has been 48 proposed that aims to link blocking responses to different features of the global warming pattern. Upper- 49 level tropical warming might be the key factor leading to a reduced blocking, because of the strengthening of 50 zonal winds (Kennedy et al., 2016). The more controversial influence of near-surface Arctic warming might 51 lead to an increased blocking frequency (Mori et al., 2014; Francis and Vavrus, 2015) (see Chapter 10, Box 52 10.1). 53 54 Figure 4.28 shows a clear decrease in blocking activity over Greenland and North Pacific for SSP7.0 and 55 SSP8.5. Models with the largest decrease in blocking frequency in boreal winter are those showing the 56 smallest frequency bias during the historical period (Davini and D’Andrea, 2020). In conclusion, there is Do Not Cite, Quote or Distribute 4-66 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 medium confidence that the frequency of atmospheric blocking events over Greenland and the North Pacific 2 will decrease in boreal winter in the SSP3-7.0 and SSP5-8.5 scenarios. 3 4 5 [START FIGURE 4.28 HERE 6 7 Figure 4.28: Projected wintertime atmospheric blocking frequencies. Box plot showing December-to-March 8 atmospheric blocking frequencies from historical simulations over 1995–2014 and projections over 2081– 9 2100, over (a) the Central European region (20°W–20°E, 45°N–65°N), (b) the Greenland region (65°W– 10 20°W, 62.5°N–72.5°N), (c) the North Pacific region (130°E–150°W, 60°N–75°N). Values show the 11 percentage of blocked days per season following the (Davini et al., 2012) index. Median values are the 12 thick black horizontal bar. The lower whiskers extend from the first quartile to the smallest value in the 13 ensemble, and the upper whiskers extend from the third quartile to the largest value. The whiskers are 14 limited to an upper bound that is 1.5 times the interquartile range (the distance between the third and first 15 quartiles). Black dots show outliers from the whiskers. The numbers below each bar report the number of 16 models included. Observationally based values are obtained as the average of the ERA-Interim 17 Reanalysis, the JRA-55 Reanalysis and the NCEP/NCAR Reanalysis. Adapted from (Davini and 18 D’Andrea, 2020). Further details on data sources and processing are available in the chapter data table 19 (Table 4.SM.1). 20 21 [END FIGURE 4.28 HERE] 22 23 24 4.5.2 Ocean 25 26 4.5.2.1 Ocean Temperature 27 28 Projections of long-term ocean thermal properties are assessed comprehensively in Chapter 9, Sections 29 9.2.1.1 and 9.2.2.1 and are not covered here to avoid unnecessary overlap. 30 31 4.5.2.2 Ocean acidification 32 33 The model-simulated long-term trend of ocean acidification is assessed in Section 4.3.2.5 and Chapter 5, 34 Section 5.3.4.1. It is virtually certain that surface ocean acidification will continue in response to the rise in 35 atmospheric CO2, and continued penetration of anthropogenic CO2 from the surface to the deep ocean will 36 acidify the ocean interior (Figure 4.29). By the end of this century, under SSP3-7.0, a pH reduction of about 37 0.3 is found at a few hundred meters depth of the global ocean, with stronger acidification in the interior 38 North Atlantic and the mid-to-high-latitude Southern Ocean. At a depth of about 1 km, a pH reduction of 39 about 0.1 is found. 40 41 Projections with CMIP6 ESMs (Kwiatkowski et al., 2020) show a surface pH decline of –0.16 ± 0.002 (±1 42 standard deviation) under SSP1-2.6 and –0.44 ± 0.005 under SSP5-8.5 from 1870–1899 to 2080–2099. The 43 high-latitude oceans, in particular the Arctic, show greater decline in pH and accelerated acidification 44 (Terhaar et al., 2020). For the same period, model-projected bottom-water pH decline is –0.018 ± 0.001 45 under SSP1-2.6 and –0.030 ± 0.002 under SSP5-8.5. The projected large scale surface ocean acidification 46 will be primarily determined by the pathway of atmospheric CO2, with weak dependence on change in 47 climate (Hurd et al., 2018) (see also Section 5.3.4.1) (high confidence). However, for a given atmospheric 48 CO2 scenario, uncertainty in projected ocean acidification increases with ocean depth because of model- 49 simulated differences in ocean circulation that transports anthropogenic CO2 from the surface to bottom 50 ocean (Kwiatkowski et al., 2020) (high confidence). For example, projected surface pH fully separates 51 between SSPs scenarios before 2050, but some overlap across SSPs is still found for projected bottom-water 52 pH in 2080 (Kwiatkowski et al., 2020). 53 54 55 [START FIGURE 4.29 HERE] 56 Do Not Cite, Quote or Distribute 4-67 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Figure 4.29: Long-term change of annual and zonal ocean pH. Displayed are multi-model mean change in annual 2 and zonal ocean pH in 2081–2100 relative to the mean of 1995–2014 for SSP1-2.6 and SSP3-7.0, 3 respectively. Eleven CMIP6 model results are used. Diagonal lines indicate regions where fewer than 4 80% of the models agree on the sign of the change and no overlay where at least 80% of the models agree 5 on the sign of change. Further details on data sources and processing are available in the chapter data 6 table (Table 4.SM.1). 7 8 [END FIGURE 4.29 HERE] 9 10 11 4.5.3 Modes of Variability 12 13 4.5.3.1 Northern and Southern Annular Modes 14 15 The Northern Annular Mode 16 AR5 assessed from CMIP5 simulations that the future boreal wintertime NAM is very likely to exhibit 17 natural variability and forced trends of similar magnitude to that observed in the historical period and is 18 likely to become slightly more positive in the future. Considerable uncertainty is related to physical 19 mechanisms to explain the observed and projected changes in the NAM, but NAM trends are clearly closely 20 connected to projected shifts in the mid-latitude jets and storm tracks. 21 22 NAM projections from climate models analysed since AR5 reveal broadly similar results the late 21st 23 century. CMIP6 models show a positive ensemble-mean trend in most seasons and the higher emission 24 scenarios that is comparable to between-model or between-realization variability (Figure 4.30a). The NAM 25 generally becomes more positive by the end of the century except in boreal summer (JJA) when there is no 26 change in the NAM in these simulations. In boreal winter (DJF) under SSP5-8.5, the central estimate is an 27 increase in the NAM by almost 3 hPa in the long-term compared to 1995–2014. This can be compared to a 28 multi-model mean interannual standard deviation in the winter NAM index of 3.4 hPa during the period 29 1850–1900. We conclude with high confidence that in the mid- to long-term, the boreal wintertime surface 30 NAM is more positive under SSP3-7.0 and SSP5-8.5, while under SSP1-1.9 and SSP1-2.6, the NAM does 31 not show any robust change. 32 33 The Southern Annular Mode 34 AR5 assessed it is likely that the evolution of the SAM over the 21st century will be primarily determined by 35 the interplay between the effects of ozone recovery and changing GHG concentrations and influence the 36 SAM in opposing ways. Owing to the relative effects of these two drivers, CMIP5 model SAM and SH 37 circulation projections differed markedly across forcing scenarios and across seasons (Barnes and Polvani, 38 2013; Barnes et al., 2014). CMIP5 models simulated a weak negative SAM trend in austral summer for 39 RCP4.5 by the end of the century (Zheng et al., 2013a), while for RCP8.5 they simulated a weak positive 40 SAM trend in austral summer (Zheng et al., 2013a). A substantial fraction of the spread in CMIP5 41 projections of the end of century SH summer jet shift under RCP8.5 may be attributable to differences in the 42 simulated change in break-up of the stratospheric polar vortex, with models that produce a later break-up 43 date showing a larger summertime poleward jet shift (Ceppi and Shepherd, 2019b). For RCP2.6, the effect of 44 ozone recovery on the SAM has been found to dominate over that of GHGs in austral summer (Eyring et al., 45 2013). In austral winter, the poleward shift of the SH circulation in CMIP5 models, and the associated 46 increase in the SAM index, tends to be larger, on average, in higher forcing scenarios though with substantial 47 inter-model spread (Barnes et al., 2014). New research since the AR5 shows that the previous theory for the 48 apparent relationship across models between the annual mean climatological SH jet position and the 49 amplitude of forced SH jet shift (Kidston and Gerber, 2010) does not hold at seasonal timescales (Simpson 50 and Polvani, 2016b). 51 52 In most seasons, the SAM becomes more positive by the end of the century relative to 1995–2014 under 53 SSP2-4.5, SSP3-7.0, and SSP5-8.5 (Figure 4.30b). Conversely, under SSP1-1.9 and SSP1-2.6, in most 54 seasons the SAM index does not show a robust change compared to 1995–2014 except in austral summer 55 when it becomes significantly more negative. The greatest change in the SAM occurs in austral winter, Do Not Cite, Quote or Distribute 4-68 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 where CMIP6 models show an ensemble-mean increase in the SAM index of almost 5 hPa in SSP5-8.5. This 2 can be compared to a multi-model mean interannual standard deviation in the austral winter SAM index of 3 4.0 hPa during 1850–1900. In conclusion, there is high confidence that in high emission scenarios (SSP3-7.0 4 and SSP5-8.5) the SAM becomes more positive in all seasons, while in the lowest scenario (SSP1-1.9) there 5 is a robust decrease in austral summer. 6 7 8 [START FIGURE 4.30 HERE] 9 10 Figure 4.30: CMIP6 Annular Mode index change from 1995–2014 to 2081–2100: (a) NAM and (b) SAM. The 11 NAM is defined as the difference in zonal mean SLP at 35°N and 65°N (Li and Wang, 2003) and the 12 SAM as the difference in zonal mean SLP at 40°S and 65°S (Gong and Wang, 1999). The shadings are 13 the 5–95% ranges across the simulations. The numbers near the top are the numbers of model simulations 14 in each SSP ensemble. Further details on data sources and processing are available in the chapter data 15 table (Table 4.SM.1). 16 17 [END FIGURE 4.30 HERE] 18 19 20 4.5.3.2 El Niño-Southern Oscillation 21 22 AR5 assessed that it is very likely that ENSO will remain the dominant mode of interannual variability in the 23 future. Moreover, due to increased moisture availability, the associated precipitation variability on regional 24 scales was assessed to likely intensify. An eastward shift in the patterns of temperature and precipitation 25 variations in the North Pacific and North America related to El Niño and La Niña teleconnections was 26 projected with medium confidence. The stability of teleconnections to other regional implications including 27 those in Central and South America, the Caribbean, parts of Africa, most of Asia, Australia and most Pacific 28 Islands were assessed to be uncertain (Christensen et al., 2013). 29 30 There is no consensus on changes in amplitude of ENSO SST variability across CMIP iterations. The main 31 factors driving the diversity of ENSO SST amplitude change in climate models are internal variability, SST 32 mean warming pattern, and model systematic biases. First, pronounced low-frequency modulations of ENSO 33 exist even in unforced control simulations due to internal variability, which leads a large uncertainty in 34 quantifying future ENSO changes (Wittenberg, 2009; Vega-Westhoff and Sriver, 2017; Zheng et al., 2018). 35 Second, ENSO characteristics depend on the climate mean state of the tropical Pacific; however, ENSO can 36 also influence the mean state through nonlinear processes (Cai et al., 2015; Timmermann et al., 2018). The 37 response of the tropical Pacific mean state to anthropogenic forcing is characterized by a faster warming on 38 the equator compared to the off-equatorial region, a faster warming of the eastern equatorial Pacific 39 compared to the central tropical Pacific (e.g., El Niño-like mean SST warming, see Chapter 7, Section 40 7.4.4.2), and a weakening of the Walker circulation in most models. Those models with a El Niño-like 41 warming tend to project a strengthening of ENSO SST variability whereas models with a La Niña-like 42 warming tend to project a weakening of variability (Zheng et al., 2016; Kohyama and Hartmann, 2017; 43 Wang et al., 2017b; Cai et al., 2018a; Fredriksen et al., 2020b). Third, how to take model biases into account 44 leads to different ENSO changes. (Kim et al., 2014) suggested that a subset of CMIP5 models that simulate 45 linear ENSO stability realistically exhibit a decrease in ENSO amplitude by the second half of the 21st 46 century. However, an increase of ENSO SST variability has been projected when considering biases in 47 ENSO pattern simulation by different models (Zheng et al., 2016; Cai et al., 2018a). This highlights the 48 importance of constraining tropical Pacific mean state changes in order to enhance confidence in the 49 projected response of ENSO. 50 51 There is also no robust consensus on changes in ENSO diversity. Several studies suggest that an increase in 52 Eastern Pacific (EP)-ENSO events tends to be projected particularly in the models with an El Niño-like 53 warming (Zheng et al., 2016; Cai et al., 2018a; Fredriksen et al., 2020a). However, Freund et al. (2020) 54 suggested that models with a El Niño-like mean warming show a tendency toward more Central Pacific (CP) 55 events but fewer EP events compared to models with an La Niña-like warming in both CMIP5 and CMIP6 56 models. Do Not Cite, Quote or Distribute 4-69 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 Even though there is limited agreement in simulated changes in ENSO SST variability, the majority of 3 models project an increase in amplitude of ENSO rainfall variability attributable to the increase in mean SST 4 and moisture in CMIP5 (Power et al., 2013; Watanabe et al., 2014; Huang and Xie, 2015) and CMIP6 (Yun 5 et al., 2021). It is likely that extreme El Niño events, accompanied by the eastern equatorial Pacific rainfall 6 exceeding the 5mm day-1 rainfall threshold, will increase in intensity (Cai et al., 2014a, 2017). However, it 7 has also been suggested that historical model biases over the equatorial Pacific cold tongue in CMIP5 may 8 lead to the greater precipitation mean change and amplification of extreme ENSO-associated rainfall in 9 CMIP5 (Stevenson et al., 2021). 10 11 There is limited intermodel agreement on future changes in ENSO teleconnections largely depending on 12 changes in the mean state and changes in ENSO properties (Yeh et al., 2018a). Many CMIP5 and CMIP6 13 models project that the centres of the extratropical teleconnection over North Pacific and North America will 14 shift eastward in association with an eastward shift in tropical convective anomalies (Yeh et al., 2018b; 15 Fredriksen et al., 2020a). There is an indication that tropical cyclones will become more frequent during 16 future El Niño events (and less frequent during future La Niña events) by the end of the 21st century (Chand 17 et al., 2017), thus contributing to the projected increase in ENSO-associated hydroclimate impacts. 18 19 While CMIP6 models show no robust change in ENSO SST amplitude in the mid- and long-term period 20 across all four SSPs, a robust increase in ENSO rainfall amplitude is found particularly in SSP2-4.5, SSP3- 21 7.0, and SSP5-8.5 (Figure 4.10). The changes in ENSO rainfall amplitude in the long-term future (2081– 22 2100) relative to the recent past (1995–2014) are statistically significant at the 95% confidence. 23 24 To conclude, the forced change in ENSO SST variability is highly uncertain in CMIP5 and CMIP6 models 25 (medium confidence). However, it is very likely that ENSO-related rainfall variability will increase 26 significantly regardless of ENSO amplitude changes in the mid- and long-term future. It is likely that the 27 pattern of ENSO teleconnection over the North Pacific and North America will shift eastward. 28 29 30 4.5.3.3 Indian Ocean Basin and Dipole Modes 31 32 In the mid- to long-term, projected climate mean state changes in the tropical Indian Ocean are expected to 33 resemble a positive IOD state, with faster warming in the west compared to the east (Cai et al., 2013; Zheng 34 et al., 2013b). However, it was argued that this projected mean state change could be due to the large mean 35 state biases in the simulated current climate and potentially not a realistic outcome (Li et al., 2016a). Mean 36 state biases also lead to lack of consensus on projected equatorial Indian Ocean SST variability and 37 equatorial modes of climate variability independent of the IOD (DiNezio et al., 2020). If mean state change 38 will indeed resemble a positive IOD state, however, this would lead to a reduction in the amplitude 39 difference between positive and negative IOD events, but with no robust change in IOD frequency (Cai et 40 al., 2013). For a small subset of CMIP5 models that simulate IOD events best, a slight increase in IOD 41 frequency was found under the CMIP5 RCP4.5 scenario (Chu et al., 2014) . 42 43 However, it was also found that the frequency of extreme positive IOD events, which exhibit the largest 44 climate impacts, might increase by a factor of about three under the CMIP5 RCP8.5 scenario (Cai et al., 45 2014b). Partially consistent with the above result, a more recent study by (Cai et al., 2021) based on CMIP5 46 RCP8.5 and CMIP6 SSP5-8.5 simulations shows a robust increased SST variability of large positive IOD 47 events, but a decreased variability of moderate IOD events. An approximate doubling of these extreme 48 positive IOD events was still found for global warming of 1.5°C warming above pre-industrial levels, 49 without a projected decline thereafter (Cai et al., 2018b). These results depend, however, on the realism of 50 the projected mean state change in the Indian Ocean (Li et al., 2016a). 51 52 To conclude, the forced change in IOD in mid- and long-term future remains uncertain due to limited lines of 53 evidence and its dependence on model mean biases. However, there is low confidence that the frequency of 54 extreme positive IOD events will increase under the high-emission scenario of SSP5-8.5. 55 Do Not Cite, Quote or Distribute 4-70 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 4.5.3.4 Tropical Atlantic Modes 3 4 AR5 assessed that there is low confidence in projected changes of the TAV because of the general failure of 5 climate models to simulate main aspects of this variability such as the northward displaced ITCZ. The 6 models that best represent the Atlantic meridional mode (AMM) show a weakening for future climate 7 conditions. However, model biases in representation of Altantic Niños strongly limit an assessment of future 8 changes. 9 10 Long-term changes in TAVs and associated teleconnections are expected as a result of global warming, but 11 large uncertainties exist due to the models’ systematic underestimation of the connection between PDV and 12 Indo-Pacific SST variations (Lübbecke et al., 2018; Cai et al., 2019b). Observational analyses show large 13 discrepancies in SST and trade winds strength (Servain et al., 2014; Mohino and Losada, 2015). Single- 14 model sensitivity experiments show that Atlantic Niño characteristics at the end of 21st century remain 15 consistent with those of the 20th century, though changes in the climatological SSTs can lead to changes in 16 the associated teleconnections (Mohino and Losada, 2015). 17 18 The weakening of the AMOC expected from global warming (see Section 4.3.2.3) has been suggested to 19 have an influence on the mean background state of tropical-Atlantic surface conditions, thereby enhancing 20 equatorial Atlantic variability and resulting in a stronger tropical Atlantic–ENSO teleconnection (Svendsen 21 et al., 2014) (see Chapter 3 Section 3.7.5 for a detailed discussion). A recent multi-model study, based on 22 CMIP5, concluded that the TAV-Pacific teleconnection will weaken under global warming due to the 23 increased thermal stability of the atmosphere (Jia et al., 2019a). However, there is still a clear lack of model 24 studies, and hence no robust evidence on the long-term evolution of TAV and associated teleconnections. 25 26 27 4.5.3.5 Pacific Decadal Variability 28 29 AR5 assessed that there is low confidence in projections of future changes in Pacific decadal variability 30 (PDV) due to the inability of CMIP5 models to represent the connection between PDV and Indo-Pacific SST 31 variations. Because the PDV appears to encompass the combined effects of different dynamical processes 32 operating at different timescales, representation of PDV in climate models remains a challenge (see Chapter 33 3, Section 3.7.6) and its long-term evolution under climate change uncertain. 34 35 In addition to uncertainty from the future evolution of the mechanisms that determined the PDV, it is also 36 unclear how the background state in the Pacific Ocean will change due to time-varying radiative forcing, and 37 how this change will interact with variability at interannual and low-frequency timescales (Fedorov et al., 38 2020). Recent research suggests that the PDV will have a weaker amplitude and higher frequency with 39 global warming (Zhang and Delworth, 2016; Xu and Hu, 2017; Geng et al., 2019). The former appears to be 40 associated with a decrease in SST variability and the meridional gradient over the Kuroshio-Oyashio region, 41 with a reduction in North Pacific wind stress and meandering of the subpolar/subtropical gyre interplay 42 (Zhang and Delworth, 2016). The latter is hypothesized to rely on the enhanced ocean stratification and 43 shallower mixed layers of a warmer climate, which would increase the phase speed of the westward- 44 propagating oceanic waves, hence shortening the decadal-interdecadal component (Goodman and Marshall, 45 1999; Zhang and Delworth, 2016; Xu and Hu, 2017). The weakening of the PDV in a warmer climate may 46 reduce the internal variability of global mean surface temperature, to which PDV seems associated (Zhang et 47 al., 1997; Kosaka and Xie, 2016; Geng et al., 2019). Thus, a weaker and higher frequency PDV could reduce 48 the contribution of internal variability to the GSAT trend and eventually lead to a reduced probability of 49 surface-warming hiatus events. 50 51 In summary, based on CMIP5, there is medium confidence that a weaker and higher frequency PDV is 52 expected under global warming. 53 54 55 4.5.3.6 Atlantic Multidecadal Variability Do Not Cite, Quote or Distribute 4-71 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 Based on paleoclimate reconstructions and model simulations, AR5 assessed that AMV is unlikely to change 3 its behaviour in the future. However, AMV fluctuations over the coming decades are likely to influence 4 regional climate, enhancing or offsetting some of the effects of global warming. 5 6 Recent proxy-derived reconstructions of AMV-related signals show persistent multi-decadal variability over 7 the last three centuries (Kilbourne et al., 2014; Svendsen et al., 2014; Moore et al., 2017), up to the last 8 millennium (Chylek et al., 2011; Zhou et al., 2016; Wang et al., 2017b) and beyond (Knudsen et al., 2011). 9 This implies that in the past AMV properties were little affected by large climatic excursions. 10 11 AMV long-term changes under future warming scenarios have been so far scarcely investigated. A study on 12 the CMIP5 multi-model simulations under RCP8.5 scenario by (Villamayor et al., 2018) found no substantial 13 differences in the simulated SST patterns (and in the related tropical rainfall response) when RCP8.5, 14 historical and piControl simulations are compared. Such results suggest that the AMV is not expected to 15 change under global warming. A more recent single-model large ensemble study (Hand et al., 2020) shows a 16 pronounced change in the AMV pattern under global warming linked to a strong reduction of the mean 17 AMOC and its variability. However, since a superposition of multiple processes controls the AMV, as 18 extensively discussed in Annex IV, Section AIV.2.7, in Chapter 3 (Section 3.7.7), and in Chapter 9 (Section 19 9.2.3.1), the length of the RCP8.5 simulations might be not sufficient to properly evaluate the respective 20 weight and interplay of internal components and influences from external forcing on AMV projections. 21 22 In conclusion, on the basis of paleoclimate reconstructions and CMIP5 model simulations, there is low 23 confidence that the AMV is not expected to change in the future. 24 25 26 4.6 Implications of Climate Policy 27 28 4.6.1 Patterns of Climate Change for Specific Levels of Global 29 30 This subsection provides an assessment of changes in climate at 1.5°C, 2°C, 3°C, and 4°C of global warming 31 relative to the period 1850–1900 (see Chapter 1, Section 1.6.2), in particular a discussion of the regional 32 patterns of change in temperature (Section 4.6.1.1), precipitation (Section 4.6.1.2), and aspects of 33 atmospheric circulation (Section 4.6.1.3). An assessment of changes in extreme weather events as a function 34 of different levels of global warming is provided in Chapter 11, while corresponding analyses of regional 35 climate change are provided in Chapter 12 and in Atlas. This section builds upon assessments from AR5 36 (Bindoff et al., 2013; Christensen et al., 2013; Collins et al., 2013; Hartmann et al., 2013) and SR1.5 (SR1.5; 37 Hoegh-Guldberg et al., 2018), as well as new literature related to projections of climate at 1.5°C, 2°C, and 38 higher levels of global warming above pre-industrial levels. 39 40 Several methodologies have been applied to estimate the spatial patterns of climate change associated with a 41 given level of global warming. These include performing model simulations under stabilisation scenarios 42 designed to achieve a specific level of global warming, the analysis of epochs identified within transient 43 simulations that systematically exceed different thresholds of global warming (e.g. Dosio et al., 2018; 44 Hoegh-Guldberg et al., 2018), Kjellström et al., 2018; Mitchell et al., 2017), and analysis based on statistical 45 methodologies that include empirical scaling relationships (ESR) (Dosio and Fischer, 2018)(Schleussner et 46 al., 2017)(Seneviratne et al., 2018) and statistical pattern scaling (e.g., Kharin et al., 2018). These different 47 methodologies are assessed in some detail in Section 4.2.5 (see also James et al., 2017) and generally 48 provide qualitatively consistent results regarding changes in the spatial patterns of temperature and rainfall 49 means and extremes (see Chapter 11) at different levels of global warming. 50 51 In this subsection, we present the projected patterns of climate change obtained following the epoch 52 approach (also called the time-shift method, see Section 4.2.4) under the Tier 1 SSPs (SSP1-2.6, SSP2-4.5, 53 SSP3-7.0 and SSP5-8.5). For each model simulation considered under each of these SSPs, 20-year moving 54 averages of the global average atmospheric surface temperature are first constructed, then this time series is 55 used to detect the first year during when GSAT exceeds the 1.5°C, 2°C, 3°C and 4°C thresholds with respect Do Not Cite, Quote or Distribute 4-72 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 to the 1850–1900 (Cross-Chapter Box 11.1). The temperature thresholds are not exceeded in all the model 2 simulations across the Tier 1 SSPs, that is, decreasing numbers of simulations are available for the analysis 3 of patterns of change as the temperature threshold increases. For each simulation within which a given 4 temperature threshold is exceeded, a 20-year global climatology is subsequently constructed to represent that 5 level of global warming, centred on the year for which the threshold was first exceeded. The composite of all 6 such climatologies across the Tier 1 SSPs and model simulations constitute the spatial patterns of change for 7 a given temperature threshold. Some of the complexities of scaling patterns of climate change with different 8 levels of global warming are also discussed in the following sections. These include overshoot versus 9 stabilization scenarios and limitations of pattern scaling for strong mitigation and stabilization scenarios 10 (Tebaldi and Arblaster, 2014). At least for the case of annual mean temperature and precipitation, strong 11 evidence exists that even for strong mitigation and stabilization scenarios, patterns of change at lower levels 12 of warming scale similarly to those reconstructed from transient simulations using either standard pattern- 13 correlation or time-shift methodologies (Tebaldi and Knutti, 2018). 14 15 Pattern scaling performance based on scenario experiments is generally better for near-surface temperature 16 than for precipitation (Ishizaki et al., 2013). For precipitation, rapid adjustments due to different forcing 17 agents must be accounted for (Richardson et al., 2016). Possible non-linear responses to different forcing 18 levels are also important (Good et al., 2015, 2016). Pattern scaling does not work as well at high forcing 19 levels (Osborn et al., 2018). It is also important to distinguish the forced response from internal variability 20 when comparing similar warming levels (Suarez-Gutierrez et al., 2018). The purpose of this section is not to 21 repeat the analysis for all the variables considered in Sections 4.4 and 4.5, but rather to show a selected 22 number of key variables that are important from the perspective of understanding the response of the 23 physical climate system to different levels of warming. 24 25 26 4.6.1.1 Temperature 27 28 Global warming of 1.5°C implies higher mean temperatures compared to 1850–1900, with generally higher 29 warming over land compared to ocean areas (virtually certain) and larger warming in high latitudes 30 compared to low latitudes (Figure 4.31). In addition, global warming of 2°C versus 1.5°C results in robust 31 increases in the mean temperatures in almost all locations, both on land and in the ocean (virtually certain), 32 with subsequent further warming at almost all locations at higher levels of global warming (virtually certain) 33 (Hoegh-Guldberg et al., 2018). For each particular level of global warming, relatively larger mean warming 34 is projected for land regions (virtually certain, see Figure 4.31; Christensen et al., 2013; Collins et al., 2013; 35 Seneviratne et al., 2016). The projected changes at 1.5°C and 2°C global warming are consistent with 36 observed historical global trends in temperature and their attribution to anthropogenic forcing (see Chapter 37 3), as well as with observed changes under the recent global warming of 0.5°C (Hoegh-Guldberg et al., 38 2018; Schleussner et al., 2017). That is, spatial patterns of temperature changes associated with the 0.5°C 39 difference in GMST warming between 1991–2010 and 1960–1979 (Schleussner et al., 2017; SR1.5) are 40 consistent with projected changes under 1.5°C and 2°C of global warming. 41 42 The largest increase in annual mean temperature is found in the high latitudes of the Northern Hemisphere 43 (NH) across all levels of global warming (virtually certain; Figure 4.31). This phenomenon peaks in the 44 Arctic and is known as Arctic amplification, with the underlying physical mechanisms assessed in detail in 45 Section 4.5.1 and Chapter 7, Section 7.4.4.1. For the CMIP6 ensemble average considered here, Arctic 46 annual mean temperatures warm by a factor of 2.3, 2.5, 2.4 and 2.4 for 1.5°C, 2°C, 3°C and 4°C of global 47 warming, respectively. That is, Arctic warming scales approximately linearly with GSAT. Generally, when 48 Arctic amplification is considered across individual models, warming occurs at a factor of 2–4 times the 49 global level of warming. It is unlikely that warming in the SH high latitudes in the 21st century will exceed 50 the change in GSAT, or that it will substantially exceed warming in the tropics, for GSAT change ranging 51 between 1.5°C and 4°C (Figure 4.31, Table 4.2). Correspondingly, there is low confidence of Antarctic 52 amplification occurring under transient, 21st century low mitigation scenarios (Table 4.2; Chapter 7, Section 53 7.4.4.1). The Antarctic continent is projected to warm at a higher rate than the mid-latitude Southern Ocean, 54 however, at all levels of global warming (Figure 4.31). The relevant physical mechanisms that reduce the 55 amplitude of polar amplification over Antarctica compared to the Arctic are assessed in detail in Section Do Not Cite, Quote or Distribute 4-73 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 4.5.1 and Chapter 7, Section 7.4.4.1. In the SH the strongest warming over land is to occur, at any given 2 level of global warming, over the subtropical areas of South America, southern Africa and Australia (high 3 confidence). The relatively strong warming in subtropical southern Africa may be attributed to strong soil- 4 moisture–temperature coupling and projected increased dryness under enhanced subsidence (Engelbrecht et 5 al., 2015; Vogel et al., 2017). Across the globe, in the tropics, subtropics, and mid- to high latitudes, 6 temperatures tend to scale linearly with the level of increase in GSAT and patterns of change are largely 7 scenario independent (high confidence). 8 9 10 [START FIGURE 4.31 HERE] 11 12 Figure 4.31: Projected spatial patterns of change in annual average near-surface temperature (°C) at different 13 levels of global warming. Displayed are (a–d) spatial patterns of change in annual average near-surface 14 temperature at 1.5°C, 2°C, 3°C, and 4°C of global warming relative to the period 1850–1900 and (e–g) 15 spatial patterns of differences in temperature change at 2°C, 3°C, and 4°C of global warming compared to 16 1.5°C of global warming. The number of models used is indicated in the top right of the maps. No 17 overlay indicates regions where the change is robust and likely emerges from internal variability, that is, 18 where at least 66% of the models show a change greater than the internal-variability threshold (see 19 Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions 20 with no change or no robust significant change, where fewer than 66% of the models show change greater 21 than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 22 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all 23 models agree on the sign of change. Values were assessed from a 20-year period at a given warming 24 level, based on model simulations under the Tier-1 SSPs of CMIP6. Further details on data sources and 25 processing are available in the chapter data table (Table 4.SM.1). 26 27 [END FIGURE 4.31 HERE] 28 29 30 4.6.1.2 Precipitation 31 32 While global mean precipitation increases as GSAT rises with the very likely range of 1–3% per 1°C (high 33 confidence, see Sections 8.2.1 and 8.4.1), patterns of precipitation change do not scale as linearly with GSAT 34 increase. Nevertheless, common features of precipitation change in the multi-model mean across scenarios 35 still exist for different levels of global warming (Figure 4.32). Precipitation will very likely increase in the 36 high latitudes and over tropical regions, and likely increase in large parts of the monsoon region, but likely 37 decrease over the subtropical regions, including the Mediterranean, southern Africa, parts of Australia and 38 South America at all four levels of global warming. The increases and decreases in precipitation will amplify 39 at higher levels of global warming (high confidence) (Figure 4.32). Changes in extreme precipitation events 40 under different levels of global warming are assessed in Chapter 11. 41 42 43 [START FIGURE 4.32 HERE] 44 45 Figure 4.32: Projected spatial patterns of change in annual average precipitation (expressed as a percentage 46 change) at different levels of global warming. Displayed are (a–d) spatial patterns of change in annual 47 precipitation at 1.5°C, 2°C, 3°C, and 4°C of global warming reletive to the period 1850–1900. No map 48 overlay indicates regions where the change is robust and likely emerges from internal variability, that is, 49 where at least 66% of the models show a change greater than the internal-variability threshold (see 50 Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions 51 with no change or no robust significant change, where fewer than 66% of the models show change greater 52 than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 53 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all 54 models agree on the sign of change. Values were assessed from a 20-year period at a given warming 55 level, based on model simulations under the Tier-1 SSPs of CMIP6. Further details on data sources and 56 processing are available in the chapter data table (Table 4.SM.1). 57 Do Not Cite, Quote or Distribute 4-74 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 [END FIGURE 4.32 HERE] 2 3 4 SR1.5 stated low confidence regarding changes in global monsoons at 1.5°C versus 2°C of global warming, 5 as well as differences in monsoon responses at 1.5°C versus 2°C. Generally, statistically significant changes 6 in regional annual average precipitation are expected at a global mean warming of 2.5°C–3°C or more 7 (Tebaldi et al., 2015). Over the Austral-winter rainfall regions of south-western South America, South Africa 8 and Australia, projected decreases in mean annual rainfall show high agreement across models and a strong 9 climate change signal even under 1.5°C of global warming, with further amplification of the signal at higher 10 levels of global warming (Mindlin et al., 2020) (high confidence). This is a signal evident in observed 11 rainfall trends over these regions (see Chapter 2, Section 2.3.1.3, and Chapter 8, Section 8.3.1.6 ). Also, over 12 the Asian monsoon regions, increases in rainfall will occur at 1.5°C and 2°C of global warming (Chevuturi et 13 al., 2018). At warming levels of 1.5°C and 2°C, the changes in global monsoons are strongly dependent on 14 the modelling strategies used, such as fully coupled transient, fully coupled quasi-equilibrium, and 15 atmosphere-only quasi-equilibrium simulations. In particular, the differences of regional monsoon changes 16 among model setups are dominated by strategy choics such as transient versus quasi-equilibrium set-up, 17 prescription of SST, and treatment of aerosols (Zhang and Zhou, 2021). 18 19 20 [START FIGURE 4.33 HERE] 21 22 Figure 4.33: Area fraction of significant precipitation change at 1.5°C, 2°C, 3°C, and 4°C of global warming. 23 Range of land fraction (top) and global area fraction (bottom) with significant precipitation increase (left- 24 hand side) and decrease (right-hand side) in the projected annual precipitation change (%) at levels of 25 global warming compared to the period 1850–1900. Values were assessed from a 20-year period at a 26 given warming level from SSP1-2.6, SSP3-7.0 and SSP5-8.5 in CMIP6. The solid line illustrates the 27 CMIP6-multi model mean and the shaded band is the 5–95% range across models that reach a given level 28 of warming. Further details on data sources and processing are available in the chapter data table (Table 29 4.SM.1). 30 31 [END FIGURE 4.33 HERE] 32 33 34 The global and land area fractions with significant precipitation changes with global warming are shown in 35 Figure 4.33. It is virtually certain that average warming will be higher over land. As warming increases, a 36 larger global and land area will experience statistically significant increases or decreases in precipitation 37 (medium confidence). The increase of the area fraction with significant precipitation increase is larger over 38 land than over the ocean, but the increase of the area fraction with significant precipitation decrease is larger 39 over the ocean than over land (Figure 4.33). Precipitation variability in most climate models increases over 40 the global land area in response to warming (Pendergrass et al., 2017). 41 42 In summary, based on the assessment of CMIP6 models, there is high confidence that global mean 43 precipitation will increase with increase in global mean surface temperature. Precipitation will very likely 44 increase in the high latitudes and over tropical regions, likely increase in large parts of the monsoon region, 45 but will likely decrease over the subtropical regions. There is high confidence that increases and decreases in 46 precipitation will amplify over higher levels of global warming . As warming increases, there is medium 47 confidence that a larger land area will experience statistically significant increases or decreases in 48 precipitation. 49 50 51 4.6.1.3 Atmospheric Circulation 52 53 AR5 reported that the application of pattern scaling to extract information on variables other than surface 54 temperature and precipitation was relatively unexplored. Since AR5, new studies have examined the 55 relationship between projections of mid-latitude atmospheric circulation and GSAT both in terms of 56 interpreting spread in responses across the CMIP5 multi-model ensemble (Grise and Polvani, 2014a, 2016) Do Not Cite, Quote or Distribute 4-75 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 and to investigate variations in the circulation response as a function of GSAT change over time within a 2 given forcing experiment (Grise and Polvani, 2017; Ceppi et al., 2018). 3 4 At a fixed time horizon, the CMIP5 multi-model spread in GSAT explains only a small fraction of the spread 5 in the shift of the NH mid-latitude circulation due to an abrupt quadrupling in CO2 (Grise and Polvani, 2016). 6 The fraction of model spread explained by GSAT in the shift of the SH circulation is larger, but still fairly 7 small (Grise and Polvani, 2014a, 2016). At a fixed time horizon and for a given emission scenario, CMIP5 8 multi-model spread in storm track shifts, and the closely related mid-latitude jets, can be better explained by 9 multi-model spread in lower and upper level meridional temperature gradients than by GSAT (Harvey et al., 10 2014; Grise and Polvani, 2016). 11 12 The transient response of the mid-latitude jets to forcing in the North Atlantic, North Pacific and Southern 13 Hemisphere behaves nonlinearly with GSAT (Grise and Polvani, 2017; Ceppi et al., 2018). This is a 14 consequence of the time-dependence of the relationship between radiative forcing and GSAT and the 15 temporal evolution of SST patterns (Ceppi et al., 2018), with a potential seasonal component in the SH 16 associated with polar stratospheric temperature changes (Grise and Polvani, 2017). Consequently, the epoch 17 approach applied to a transient simulation of the 21st century will overestimate the mid-latitude circulation 18 response in a stabilized climate. Dedicated time slice experiments simulating stabilized climates are therefore 19 required to assess differences in mid-latitude circulation at given levels of global warming (Li et al., 2018). 20 A further complication in the SH is the competing influences of ozone recovery and increasing GHG 21 concentrations on the austral-summer mid-latitude circulation during the first half the 21st century (Barnes 22 and Polvani, 2013; Barnes et al., 2014). Using transient 21st century experiments to diagnose changes in SH 23 mid-latitude circulation at different levels of warming therefore confounds the effects of ozone recovery and 24 GHG increases (Ceppi et al., 2018). Given these various limitations, we do not apply epoch analysis to 25 assess mid-latitude atmospheric circulation changes and related annular modes of variability. 26 27 28 4.6.2 Climate Goals, Overshoot, and Path-Dependence 29 30 Many scenarios aiming at limiting warming by 2100 to 1.5°C involve overshoot – ERF temporarily exceeds 31 a certain level before peaking and declining again (see also Annex VII: Glossary). To quantify the 32 implications of any such overshoot, this subsection assesses reversibility of climate due to temporary 33 overshoot of GSAT levels during the 21st century, and implications for the use of carbon budgets. It also 34 assesses differences in climate outcomes under different pathways, with a focus on comparing the SSPs used 35 in CMIP6 with the RCPs used in CMIP5. 36 37 38 4.6.2.1 Climate change under overshoot 39 40 The SR1.5 (IPCC, 2018) concluded with high confidence that overshoot trajectories ‘result in higher impacts 41 and associated challenges compared to pathways that limit global warming to 1.5°C with no or limited 42 overshoot’. The degree and duration of overshoot affects the risks and impacts likely to be experienced 43 (Hoegh-Guldberg et al., 2018) and the emissions pathway required to achieve it (Akimoto et al., 2018). 44 Consequences such as on ice sheets and climatic extremes have been found to be greater at 2°C of global 45 warming than at 1.5°C (Schleussner et al., 2016; Hoegh-Guldberg et al., 2018) but even on recovery to lower 46 temperatures, these effects may not reverse. Overshoot has been found to lead to irreversible changes in 47 thermosteric sea-level (Tokarska and Zickfeld, 2015; Palter et al., 2018; Tachiiri et al., 2019), AMOC (Palter 48 et al., 2018), ice-sheets, and permafrost carbon (Section 4.7.2, Chapter 5, Section 5.4.9) and to long-lasting 49 effects on ocean heat (Tsutsui et al., 2006a). Abrupt changes and tipping points are not well understood, but 50 the higher the warming level and the longer the duration of overshoot, the greater the risk of unexpected 51 changes (see sections 4.7.2). Non-reversal of the hydrological cycle has also been found in some studies with 52 an increase in global precipitation following CO2 decrease being attributed to a build-up of ocean heat (Wu 53 et al., 2010), and to a fast atmospheric adjustment to CO2 radiative forcing (Cao et al., 2011a). 54 55 Global temperature is expected to remain approximately constant if emissions of CO2 were to cease (Section Do Not Cite, Quote or Distribute 4-76 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 4.7.1.1), and so reductions in GSAT are only possible in the event of net negative global CO2 emissions. We 2 assess here results from an overshoot scenario (SSP5-3.4-OS; O’Neill et al., 2016), which explores the 3 implications of a peak and decline in forcing during the 21st century. Reversibility under more extreme and 4 idealised carbon dioxide removal (CDR) scenarios is assessed in Section 4.6.3. In SSP5-3.4-OS, CO2 peaks 5 at 571 ppm in the year 2062 and reverts to 497 ppm by 2100 – approximately the same level as in 2040. 6 SSP5-3.4-OS has strong net negative emissions of CO2, exceeding those in SSP1-2.6 and SSP1-1.9 from 7 2070 onwards and reaching –5.5 PgC yr-1 (–20 GtCO2 yr-1) by 2100. While this causes global mean 8 temperature to decline, changes in climate have not fully reversed by 2100 under this reversal of CO2 9 concentration (Figure 4.34). Quantities are compared for 2081–2100 relative to a 20-year period (2034– 10 2053) of the same average CO2. Differences between these two periods of the same CO2 are: GSAT: 11 0.28±0.30°C (mean ± standard deviation); global land precipitation: 0.026±0.011 mm/day; September Arctic 12 sea-ice area: –0.32±0.53 million km2; thermosteric sea-level: 12±0.8 cm. As assessed in Section 9.3.1.1, 13 Arctic sea-ice area is linearly reversible with GSAT. Although these climate quantities are not fully 14 reversible, the overshoot scenario results in reduced climate change compared with stabilisation or continued 15 increase in greenhouse gases (Tsutsui et al., 2006b; Palter et al., 2018; Tachiiri et al., 2019) (high 16 confidence). 17 18 19 [START FIGURE 4.34 HERE] 20 21 Figure 4.34: Simulated changes in climate indices for SSP5-3.4-OS plotted against atmospheric CO2 22 concentration (ppm) from 480 up to 571 and back to 496 by 2100. (a) Global surface air temperature 23 change; (b) Global land precipitation change; (c) September Arctic sea-ice area change; (d) Global 24 thermosteric sea-level change. Plotted changes are relative to the 2034–2053 mean which has same CO2 25 as 2081–2100 mean (shaded grey bar). Red lines denote changes during the period up to 2062 when CO2 26 is rising, blue lines denote changes after 2062 when CO2 is decreasing again. Thick line is multi model 27 mean; thin lines and shading show individual models and complete model range. Numbers in square 28 brackets indicate number of models used in each panel. Further details on data sources and processing are 29 available in the chapter data table (Table 4.SM.1). 30 31 [END FIGURE 4.34 HERE] 32 33 34 The transient climate response to cumulative carbon (CO2) emissions, TCRE, allows climate policy goals to 35 be associated with remaining carbon budgets as global temperature increase is near-linear with cumulative 36 emissions (Section 5.5). Research since AR5 has shown that the concept of near-linearity of climate change 37 to cumulative carbon emissions holds for measures other than just GSAT, such as regional climate (Leduc et 38 al., 2016) or extremes (Harrington et al., 2016)(Seneviratne et al., 2016). However, ocean heat and carbon 39 uptake do exhibit path dependence, leading to deviation from the TCRE relationship for levels of overshoot 40 above 300 PgC (Zickfeld et al., 2016; Tokarska et al., 2019). Sea-level rise, loss of ice-sheets, and 41 permafrost carbon release may not reverse under overshoot and recovery of GSAT and cumulative emissions 42 (Section 4.7). TCRE remains a valuable concept to assess climate policy goals and how to achieve them but 43 given the non-reversibility of different climate metrics with CO2 and GSAT reductions, it has limitations 44 associated with evaluating the climate response under overshoot scenarios and CO2 removal (medium 45 confidence). 46 47 48 4.6.2.2 Consistency between Shared Socioeconomic Pathways and Representative Concentration 49 Pathways 50 51 As CMIP5 and CMIP6 employed different scenario sets (RCPs and SSPs, respectively – see Section 1.6.1.1, 52 Cross-Chapter Box 1.4), we assess how much of the differences in projections are due to the scenario change 53 and how much due to model changes. CMIP6-simulated GSAT increases tend to be larger than in CMIP5, 54 for nominally comparable scenarios ((Tebaldi et al., 2021), see Section 4.3.1). 55 56 The radiative forcing labels on SSP and RCP scenarios is approximate and enables the multiple climate Do Not Cite, Quote or Distribute 4-77 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 forcings within the scenario to be characterised by a single number. While the scenarios are similar in terms 2 of the stratospheric adjusted radiative forcing (Tebaldi et al., 2021), they differ more in their effective 3 radiative forcing (ERF). The combination of component forcings (CO2, non-CO2 greenhouse gases, aerosols) 4 within the scenario also differ (Meinshausen et al., 2020). The ERF levels in the RCP and SSP scenarios 5 have been calculated by sampling uncertainty in forcing from a range of different GHG species and aerosols 6 (see 7.SM.1.4 for details). Figure 4.35 shows the time evolution and 2081–2100 mean across the families of 7 scenarios and how this affects projections of GSAT. That the ERFs differ between corresponding SSP and 8 RCP scenarios makes comparison between CMIP6 and CMIP5 projections challenging (Tebaldi et al., 2021). 9 (Wyser et al., 2020) find the EC-Earth3-Veg model exhibits stronger radiative forcing and substantially 10 greater warming under SSP5-8.5 than RCP8.5, and similar, but smaller additional warmings for SSP2-4.5 11 and SSP1-2.6 compared with RCP4.5 and RCP2.6, respectively. In addition to the global response, climate 12 can vary regionally due to non-CO2 components of forcing (Samset et al., 2016; Richardson et al., 2018b, 13 2018a). 14 15 Emulators (Cross-Chapter Box 7.1) can be used to aid understanding of differences between generations of 16 scenarios. AR5 (Collins et al., 2013) explored the differences between CMIP3 and CMIP5 (their Figure 17 12.40). Here we use an emulator calibrated to AR6 assessed GSAT ranges, thus eliminating the effect of 18 differences in the model ensembles, to analyse the differences between SSP and RCP scenarios. 19 MAGICC7.5 in its WGIII-calibrated setup (see Cross Chapter Box 7.1) projects differences in 2081–2100 20 mean warming between the RCP2.6 and SSP1-2.6 scenarios of around 0.2°C, between RCP4.5 and SSP2-4.5 21 of around 0.3°C and between RCP8.5 and SSP5-8.5 of around 0.3°C (Figure 4.35b). The SSP scenarios also 22 have a wider 5–95% range simulated by MAGICC7.5 explaining about half of the increased range seen 23 when comparing CMIP5 and CMIP6 models. Higher climate sensitivity is, though, the primary reason 24 behind the upper end of the warming for SSP5-8.5 reaching 1.5°C higher than the CMIP5 results. Compared 25 with the differences between the CMIP5 and CMIP6 multi-model ensembles for the same scenario pairs 26 (Table A6 in Tebaldi et al., 2021), the higher ERFs of the SSP scenarios contribute approximately half of the 27 warmer CMIP6 SSP outcomes (medium confidence). 28 29 In summary, there is medium confidence that about half of the warming increase in CMIP6 compared to 30 CMIP5 is due to higher climate sensitivity in CMIP6 models; the other half arises from higher ERF in 31 nominally comparable scenarios (e.g., RCP8.5 and SSP5-8.5). 32 33 34 [START FIGURE 4.35 HERE] 35 36 Figure 4.35: Comparison of RCPs and SSPs run by a single emulator to estimate scenario differences. Time 37 series with 5–95% ranges and medians of (a) effective radiative forcings, calculated as described in 38 Annex 7.A.1; and (b) GSAT projections relative to 1850–1900 for the RCP and SSP scenarios from 39 MAGICC 7.5. Note that the nameplate radiative forcing level refers to stratospheric adjusted radiative 40 forcings in AR5-consistent settings (Tebaldi et al., 2021) while ERFs may differ. MAGICC7.5 is here run 41 in the recommended setup for WGIII, prescribing observed GHG concentrations for the historical period 42 and switching to emission-driven runs in 2015. Further details on data sources and processing are 43 available in the chapter data table (Table 4.SM.1). 44 45 [END FIGURE 4.35 HERE] 46 47 48 4.6.3 Climate Response to mitigation, Carbon Dioxide Removal, and Solar Radiation Modification 49 50 Most strong-mitigation scenarios assume – in addition to emissions reductions – some form of carbon 51 dioxide removal (CDR) – anthropogenic activities that remove CO2 from the atmosphere and durably store it 52 in geological, terrestrial, or ocean reservoirs, or in products (see Annex VII: Glossary). SR1.5 (Rogelj et al., 53 2018b) assessed that all pathways that limit warming to 1.5°C by 2100 with no or limited overshoot use 54 CDR. In the SSP class of scenarios, SSP1-1.9 is characterized by a rapid decline of net CO2 emissions to 55 zero by 2050 and net negative CO2 emissions in the second half of this century (O’Neill et al., 2016; Rogelj 56 et al., 2018a), implying the use of CDR. The term ‘net CO2 emissions’ refers to the difference between Do Not Cite, Quote or Distribute 4-78 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 anthropogenic CO2 emissions and removal by CDR options, and ‘net negative CO2 emissions’ imply a 2 scenario where CO2 removal exceeds emissions (van Vuuren et al., 2011) (van Vuuren et al., 2016). The 3 terms ‘negative emissions’ and ‘net negative emissions’ refer to and include all GHGs (see Annex VII: 4 Glossary). 5 6 Climate change can be also offset by solar radiation modification (SRM) measures that modify the Earth’s 7 radiation budget to reduce global warming (see Annex VII: Glossary). CDR and SRM approaches have been 8 together referred to as ‘geoengineering’ or ‘climate engineering’ in the literature (The Royal Society, 2009; 9 NRC, 2015a, 2015b; Schäfer et al., 2015). However, following SR1.5 (de Coninck et al., 2018), these terms 10 inconsistently used in the literature, so that CDR and SRM are explicitly differentiated here. SRM contrasts 11 with climate mitigation because it introduces a ‘mask’ to the climate change problem by altering the Earth’s 12 radiation budget, rather than attempting to address the root cause of the problem, which is the increase in 13 GHGs in the atmosphere. 14 15 Section 4.6.3.1 assesses the emergence of the climate response to mitigation, which is reflected by the 16 difference between high- and low-emission scenarios. Section 4.6.3.2 then assesses the climate response to 17 mitigation through CDR options, usually assumed against the background of some emission scenario; note 18 that the CDR options themselves are assessed in Chapter 5 (Section 5.6.2). Section 4.6.3.3 assesses the 19 climate system response to SRM options. The biogeochemical implications of CDR and SRM are assessed in 20 Chapter 5 (Sections 5.6.2 and 5.6.3, respectively). The importance of CDR for reaching net zero or negative 21 CO2 emissions in mitigation pathways is assessed in the AR6 WGIII report (Chapters 3, 4, 6, 7 and 12). The 22 risks for and impacts on human and natural systems due to SRM are assessed in the AR6 WGII report 23 (Chapter 16), and the international governance issues related to SRM and CDR are assessed in the AR6 24 WGIII report (Chapter 14). 25 26 27 4.6.3.1 Emergence of the climate response to mitigation 28 29 Reducing GHG emissions will eventually slow and limit the degree of climate change relative to high- 30 emission scenarios such as SSP5-8.5 (very high confidence). Even when CO2 emissions are reduced, 31 however, atmospheric CO2 concentrations continue to increase as long as emissions exceed removal by sinks 32 (Millar et al., 2017). Surface warming would likewise initially continue under scenarios of decreasing 33 emissions, resulting in a substantial lag between a peak in CO2 emissions and peak warming (Ricke and 34 Caldeira, 2014; Zickfeld and Herrington, 2015) (high confidence). The lag between peak emissions and 35 warming depends on the emissions history prior to the peak and also on the rate of the subsequent emissions 36 reductions (Matthews, 2010; Ricke and Caldeira, 2014; Zickfeld and Herrington, 2015). 37 38 In addition to the lag between peak emissions and peak warming, the climate response to reduced emissions 39 would be overlain by internal variability, which can amplify or attenuate the forced response. The resulting 40 masking of differences between scenarios is illustrated in Figure 4.36 for GSAT trends over 2021–2040 41 (Maher et al., 2020). The overall trends conform to expectations in that most simulations show warming 42 almost everywhere, especially under scenario RCP8.5 (Figure 4.36 bottom row). But any individual grid 43 point can in principle show no warming or even cooling, even under RCP 8.5, over the near term (Figure 44 4.36 middle row). The magnitude of pointwise maximum and minimum temperature trends can be as large as 45 0.5°C per year (Figure 4.36 top and middle rows), exceeding possible trends in the global mean by one order 46 of magnitude. While it is only a small fraction of the surface that simultaneously can show cooling, cooling 47 at any given location is fully consistent with globally averaged surface warming over the near term (high 48 confidence, since the findings of (Maher et al., 2020) are consistent across six different large initial-condition 49 ensembles). 50 51 52 [START FIGURE 4.36 HERE] 53 54 Figure 4.36: Masking of climate response to mitigation by internal variability in the near term. Near-term 55 (2021–2040) pointwise maximum (top row) and pointwise minimum (middle row) surface air Do Not Cite, Quote or Distribute 4-79 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 temperature trends in the large initial-condition ensemble from MPI (left and centre columns), and CESM 2 (right column) models in the RCP2.6 (left column) and RCP8.5 scenarios (centre and right columns). The 3 percentage of ensemble members with a warming trend in the near term is shown in the bottom panels. 4 Figure modified from (Maher et al., 2020). Further details on data sources and processing are available in 5 the chapter data table (Table 4.SM.1). 6 7 [END FIGURE 4.36 HERE] 8 9 10 An important development since AR5 has been the quantification of when the climate response to mitigation 11 can be expected to emerge from the background noise of internal variability (illustrated in Figure 4.36; see 12 Section 1.4.2.2, see also Annex VII: Glossary). A basic ambiguity arises because once mitigation measures 13 are in place, it is no longer possible to observe what the climate would have been without these measures, 14 and any statement about emergence of the response to mitigation is contingent upon the assumed strength of 15 mitigation in relation to an assumed (‘counterfactual’) no-mitigation scenario. Still, there is high agreement 16 on the emergence of the climate response to mitigation across a number of independent studies using 17 different models and different statistical approaches. 18 19 Among global quantities, emergence of the response to differing CO2 emissions – representing differences 20 between low- and high-emission scenarios – is first expected to arise in global-mean CO2 concentrations, 21 about 10 years after emission pathways have started diverging (Tebaldi and Friedlingstein, 2013; Peters et 22 al., 2017; Schwartzman and Keeling, 2020; Spring et al., 2020) (high confidence). In these studies, 23 emergence is generally defined as the time at which the global mean concentration first differs between 24 mitigation and non-mitigation scenarios by more than two standard deviations of internal variability, 25 although there are some methodological differences. 26 27 Emergence in GSAT would be delayed further, owing to the inertia in the climate system. Although not 28 investigating emergence as defined here in AR6, (Tebaldi et al., 2021) used 20-year running-mean GSAT 29 and compared pairwise either model-by-model or between CM IP6 ensemble means from the core set of five 30 scenarios assessed in this chapter. Differences by more than 0.1°C showed up in most cases in the near 31 term, with only some of the individual models and the comparisons of the closest scenarios showing a delay 32 until the mid-term. Taking internal variability explicitly into account, (Tebaldi and Friedlingstein, 2013) and 33 (Samset et al., 2020) found emergence of mitigation benefits in GSAT changes about 25–30 years after 34 RCP2.6 emissions diverge from the higher-emissions trajectories in RCP4.5 and RCP8.5. Consistently, 35 (Marotzke, 2019) found about one-third likelihood that a trend reduction in GSAT, over the period 2021– 36 2035 relative to 2005–2020, would be attributable to the emissions reductions implied by the difference 37 between RCP2.6 and RCP4.5. Emergence of the GSAT response to mitigation of individual short-lived 38 climate forcers (SLCFs) would likewise not occur until several decades after emissions trajectories diverge, 39 owing to the relatively small influence of individual SLCFs on the total ERF (Samset et al., 2020), see also 40 Section 4.4.4 and Figure 4.18. 41 42 In contrast to the earlier studies, emergence in GSAT within the near- term has recently been found by 43 (McKenna et al., 2021) who investigated the likelihood that under the SSP scenarios GSAT trends will 44 exceed the largest historical observed 20-year trends. They found that under scenario SSP1-1.9, the 20-year 45 GSAT trends would likely be lower than in SSP3-7.0 and SSP5-8.5 within the near term. This earlier 46 diagnosed time of emergence compared to (Marotzke, 2019), while using a similar statistical approach, 47 presumably arose because of the longer-period trends (20 rather than 15 years) and the larger difference 48 between emissions trajectories considered (medium confidence). Using 20-year temperature anomalies 49 relative to 1995–2014 instead of 20-year trends yielded a low probability of emergence (McKenna et al., 50 2021), consistent with the AR5 (Collins et al., 2013; Kirtman et al., 2013), (Tebaldi and Friedlingstein, 2013) 51 (Samset et al., 2020). It is not yet understood why GSAT trends appear to show faster emergence of 52 mitigation benefits, compared to GSAT anomalies. 53 54 Emergence of mitigation benefits has been studied much less for quantities other than globally and annually 55 averaged CO2 concentration and surface temperature. Boreal-winter temperatures are more challenging for 56 emergence, due to larger variability in boreal winter and adding a decade to the time of emergence, whereas Do Not Cite, Quote or Distribute 4-80 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 emergence times for boreal-summer averages are similar to the annual temperature averages (Tebaldi and 2 Friedlingstein, 2013). Emergence happens later at the regional scale, with a median time of emergence of 3 30–45 years after emission paths separate in RCP2.6 relative to RCP4.5 and RCP8.5; a stricter requirement 4 of 95% confidence level instead of median induces a delay of several decades, bringing time of emergence 5 toward the end of the 21st century at regional scales (Tebaldi and Friedlingstein, 2013). 6 7 Attribution to emissions reductions, for the case of RCP2.6 relative to RCP4.5, is not substantially more 8 likely for 2021–2035 trends in upper-2000m OHC than for GSAT (Marotzke, 2019), although OHC change 9 is thought to be less susceptible to internal variability. Furthermore, (Marotzke, 2019) found only around 10 10% likelihood of mitigation-benefit emergence during 2021–2035 for change in AMOC and September 11 Arctic sea-ice area. (Tebaldi and Wehner, 2018) showed that the differences in temperature extremes 12 between RCP4.5 and RCP8.5 over all land areas become statistically significant by 2050. The seemingly 13 contrasting result of (Ciavarella et al., 2017) that mitigation benefits arise earlier for climate extremes poses 14 no contradiction, because (Ciavarella et al., 2017) did not look at emergence as defined here but at the 15 extremes of a distribution, which differ between scenarios already at a time when the distributions are still 16 largely overlapping. 17 18 In summary, if strong mitigation is applied from 2020 onward as reflected in SSP1-1.9, its effect on 20-year 19 trends in GSAT would likely emerge during the near term, measured against an assumed non-mitigation 20 scenario such as SSP3-7.0 and SSP5-8.5. However, the response of many other climate quantities to 21 mitigation would be largely masked by internal variability during the near term, especially on the regional 22 scale (high confidence). The mitigation benefits for these quantities would emerge only later during the 21st 23 century (high confidence). During the near term, a small fraction of the surface can show cooling under all 24 scenarios assessed here, so near-term cooling at any given location is fully consistent with globally averaged 25 surface warming (high confidence). 26 27 28 4.6.3.2 Climate Response to Mitigation by Carbon Dioxide Removal 29 30 CDR options include afforestation, soil carbon sequestration, bioenergy with carbon capture and storage 31 (BECCS), wet land restoration, ocean fertilization, ocean alkalinisation, enhanced terrestrial weathering and 32 direct air capture and storage (see Chapter 5, Section 5.6.2 and Table 5.9 for a more complete discussion). 33 Chapter 8 (Section 8.4.3) assesses the implications of CDR for water cycle changes. The potential of 34 different CDR options in terms of the amount of CO2 removed per year from the atmosphere, costs, co- 35 benefits and side effects of the CDR approaches are assessed in the SR1.5 (de Coninck et al., 2018), the AR6 36 WGIII report (see Chapters 7 and 12), and in several review papers (Fuss et al., 2018; Lawrence et al., 2018; 37 Nemet et al., 2018). In the literature, CDR options are also referred to as ‘negative CO2 emission 38 technologies’. 39 40 Deployment of CDR will lead to a reduction in atmospheric CO2 levels only if uptake by sinks exceeds net 41 CO2 emissions. Hence, there could be a substantial delay between the initiation of CDR and net CO2 42 emissions turning negative (van Vuuren et al., 2016), and the time to reach net negative CO2 emissions and 43 the evolution of atmospheric CO2 and climate thereafter would depend on the combined pathways of 44 anthropogenic CO2 emissions, CDR, and natural sinks. The cooling (or avoided warming) due to CDR would 45 be proportional to the cumulative amount of CO2 removed from the atmosphere by CDR (Tokarska and 46 Zickfeld, 2015; Zickfeld et al., 2016), as implied by the near-linear relationship between cumulative carbon 47 emissions and GSAT change (see Section 5.5) 48 49 Emission pathways that limit globally averaged warming to 1.5°C or 2°C by the year 2100 assume the use of 50 CDR approaches in combination with emission reductions to follow net negative CO2 emissions trajectory in 51 the second half of this century. For instance, in SR1.5, all analysed pathways limiting warming to 1.5°C by 52 2100 with no or limited overshoot include the use of CDR to some extent to offset anthropogenic CO2 53 emissions and the median of CO2 removal across all scenarios was 730 GtCO2 in the 21st century (Rogelj et 54 al., 2018b) (Rickels et al., 2018). Affordable and environmentally and socially acceptable CDR options at 55 scale well before 2050 are an important element of 1.5°C-consistent pathways especially in overshoot Do Not Cite, Quote or Distribute 4-81 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 scenarios (de Coninck et al., 2018). The required scale of removal by CDR can vary from 1–2 GtCO2 per 2 year from 2050 onwards to as much as 20 GtCO2 p`er year (Waisman et al., 2019). In the SSP class of 3 scenarios, net CO2 emissions turn negative from around 2050 in SSP1-1.9 and around 2070 in SSP1-2.6 and 4 in the overshoot scenario SSP5-3.4-OS (O’Neill et al., 2016). Thus, CDR would play a pivotal role in 5 limiting climate warming to 1.5°C or 2°C (Minx et al., 2018). In stark contrast, however, two extensive 6 reviews (Lawrence et al., 2018; Nemet et al., 2018) conclude that it is implausible that any CDR technique 7 can be implemented at scale that is needed by 2050. 8 9 When CDR is applied continuously and at scales as large as currently deemed possible, under RCP8.5 as the 10 background scenario, the widely discussed CDR options such as afforestation, ocean iron fertilization and 11 surface ocean alkalinisation are individually expected to be relatively ineffective, with limited (8%) warming 12 reductions relative to the scenario with no CDR option (Keller et al., 2014). Hence, the potential role that 13 CDR will play in lowering the temperature in high-emission scenarios is limited (medium confidence). The 14 challenges involved in comparing the climatic effects of various CDR options has also been recognized in 15 recent studies (Sonntag, 2018; Mengis et al., 2019). For instance, due to compensating processes such as 16 biogeophysical effects of afforestation (warming from albedo decrease when croplands are converted to 17 forests) more carbon is expected to be removed from the atmosphere by afforestation than by ocean 18 alkalinisation to reach the same global mean cooling. 19 20 21 [START FIGURE 4.37 HERE] 22 23 Figure 4.37: Delayed climate response to CDR-caused net negative CO2 emissions. Multi-model simulated 24 response in global and annual mean climate variables for a ramp-up followed by ramp-down of CO2. 25 Atmospheric CO2 increases from the pre-industrial level at a rate of 1% yr-1 to 4 • CO2, then decreases at 26 the same rate to the pre-industrial level and then remains constant. The ramp-down phase represents the 27 period of net negative CO2 emissions. a) normalized ensemble mean anomaly of key variables as a 28 function of year, including atmospheric CO2, surface air temperature, precipitation, thermosteric sea-level 29 rise (see Glossary), global sea-ice area, Northern Hemisphere sea-ice area in September, and Atlantic 30 meridional overturning circulation (AMOC); b) surface air temperature; c) precipitation; d) September 31 Arctic sea-ice area; e) AMOC; f) thermostatic sea level; 5-year running means are shown for all variables 32 except the sea-level rise. In b–f, red lines represent the phase of CO2 ramp-up, blue lines represent the 33 phase of CO2 ramp-down, brown lines represent the period after CO2 has returned to pre-industrial level, 34 and black lines represent the multi-model mean. For all of the segments in b–f, the solid coloured lines 35 are CMIP6 models, and the dashed lines are other models (i.e., EMICs, CMIP5 era models). Vertical 36 dashed lines indicate peak CO2 and when CO2 again reaches pre-industrial value. The number of CMIP6 37 and non-CMIP6 models used is indicated in each panel. The time series for the multi-model means (b–f) 38 and the normalized anomalies (a) are terminated when data from all models are not available, in order to 39 avoid the discontinuity in the time series. Further details on data sources and processing are available in 40 the chapter data table (Table 4.SM.1). 41 42 [END FIGURE 4.37 HERE] 43 44 45 The climate response to CDR-caused net negative CO2 emissions has been studied in Earth system models 46 by prescribing idealized ramp-down of CO2 concentrations (Zickfeld et al., 2016) (MacDougall, 2013a) 47 (Schwinger and Tjiputra, 2018), CO2 concentrations of RCP scenarios that have net negative CO2 emissions 48 (Jones et al., 2016c), and idealized net negative CO2 emission scenarios (Tokarska and Zickfeld, 2015). The 49 Carbon Dioxide Removal Model Intercomparison Project (CDRMIP) uses multiple ESMs to explore the 50 climate response, effectiveness of CO2 removal, and challenges of CDR options (Keller et al., 2018). 51 Idealized CDRMIP simulations increase CO2 concentrations at 1% per year from the level in the pre- 52 industrial control run (piControl) to 4 • CO2 and subsequently decrease at the same rate to the piControl 53 level. This section assesses the lag in climate response to CDR-caused negative emission; climate 54 ‘reversibility’ is assessed in Section 4.7.2. The ramp-down phase, though unrealistic, represents the ‘net 55 negative CO2 emission’ phase. 56 Do Not Cite, Quote or Distribute 4-82 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Figure 4.37: illustrates the first results from CDRMIP (Keller et al., 2018). Other studies that use similar 2 (Zickfeld et al., 2016) (Schwinger and Tjiputra, 2018)(Jeltsch-Thömmes et al., 2020) or other idealized 3 scenarios (MacDougall, 2013a) or more realistic net negative CO2 emission scenarios such as RCP2.6 (Jones 4 et al., 2016c) and scenarios that limit warming to 2°C or less after different levels of overshoot (Tokarska 5 and Zickfeld, 2015) arrive at similar conclusions. Changes in key climate variables substantially lag behind 6 the decline in CO2 (Figure 4.37). The precipitation increase at the beginning of the ramp-down phase agrees 7 with the increase in precipitation for an abrupt decline in CO2 (Cao et al., 2011b). Notwithstanding a decline 8 in atmospheric CO2, global mean thermosteric sea level would continue to rise. When atmospheric CO2 9 returns to the piControl level, global mean thermosteric sea level is higher than its value at peak CO2 (Figure 10 4.37), and it is likely that thermosteric global sea level would not return to piControl levels for over 1000 11 years after atmospheric CO2 is restored to piControl concentrations (Tokarska and Zickfeld, 2015; Ehlert and 12 Zickfeld, 2018). Therefore, there is high confidence that sea-level rise will not be reversed by CDR at least 13 for several centuries (see also Chapter 9, Section 9.6.3.5). A comparison of different models shows recovery 14 of AMOC intensity during net negative CO2 emissions, but the results are model dependent – strengthening 15 with an overshoot in most models (Jackson et al., 2014;) and strengthening but not reaching the initial state 16 in some models (Sgubin et al., 2015). The overall lag in response is qualitatively similar to the lagged 17 climate system response in the overshoot scenario SSP5-34-OS where CO2 rises until 2062 and decreases 18 thereafter (Figure 4.34) The lag in climate response to CDR causes hysteresis between key climate variables 19 such as temperature, precipitation, AMOC and sea level, and atmosphere CO2 with the hysteresis 20 characteristics dependent on the rate of CDR and climate sensitivity (MacDougall, 2013b) (Jeltsch-Thömmes 21 et al., 2020). 22 23 Termination of CDR refers to a sudden and sustained discontinuation of CDR deployment (see Section 24 4.6.3.3 for termination effects of SRM). The literature on the termination effects of CDR is limited, mostly 25 considering scenarios where CDR implementation is explicit and does not result in net negative CO2 26 emissions (Keller et al., 2014; González et al., 2018). In simulations where CDR is applied on the RCP8.5 27 scenario at scales as large as currently deemed possible, the increase in global mean warming rates following 28 CDR termination are relatively small in comparison to SRM termination (Keller et al., 2014). The exception 29 is artificial ocean upwelling where surface cooling is mainly caused by bringing cold water from the deep 30 ocean; upon termination this causes larger rates of surface warming (Oschlies et al., 2010). When 31 background emissions are as high as in RCP8.5, termination of a large global-scale application of CDR such 32 as ocean alkalinisation for multiple decades could also result in large regional warming rates (up to 0.15°C 33 per year) that are comparable to those caused by termination of SRM (González et al. 2018). In such cases, 34 large amounts of CO2 would be removed from the atmosphere before termination, and termination would 35 cause a temporal trajectory of atmospheric CO2 that is parallel to the high-emission scenario but from an 36 atmosphere with much lower CO2 levels. Because CO2 radiative forcing is a logarithmic function of CO2 37 concentration, large regional warming rates are simulated in such terminations. Thus, there is high 38 confidence that the climate effect of CDR termination would depend on the amount CO2 removed by CDR 39 prior to termination and the rate of background CO2 emissions at the time of termination. See also Chapter 5, 40 Table 5.9 that summarizes the termination effects of individual CDR options. 41 42 In summary, there is high confidence that, due to the near-linear relationship between cumulative carbon 43 emissions and GSAT change, cooling or avoided warming due to a CDR option would depend on the 44 cumulative amount of CO2 removed by that CDR option. The climate system response to the deployment of 45 CDR is expected to be delayed by years (e.g., in temperature, precipitation, sea-ice extent) to centuries (e.g., 46 sea level and AMOC) (high confidence). The climate response to a sudden and sustained CDR termination 47 would depend on the amount of CDR-induced cooling prior to termination and the rate of background CO2 48 emissions at the time of termination (high confidence). 49 50 51 4.6.3.3 Climate Response to Solar Radiation Modification 52 53 Most SRM approaches, including stratospheric aerosol injection (SAI), marine cloud brightening (MCB), 54 and surface albedo enhancements (Table 4.7), aim to cool the Earth by deflecting more solar radiation to 55 space. Although cirrus cloud thinning (CCT) aims to cool the planet by increasing the longwave emission to Do Not Cite, Quote or Distribute 4-83 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 space, it is included in the portfolio of SRM options (Table 4.7) for consistency with AR5 (Boucher et al., 2 2013) and SR1.5 (de Coninck et al., 2018). Other approaches such as injection of sulphate aerosols into the 3 Arctic troposphere and sea-ice albedo enhancements for moderating regional warming have also been 4 suggested (MacCracken, 2016) (Field et al., 2018). As noted in SR1.5 (de Coninck et al., 2018), SRM is only 5 considered as a potential supplement to deep mitigation, for example in overshoot scenarios (MacMartin et 6 al., 2018). 7 8 9 [START TABLE 4.7 HERE] 10 11 Table 4.7: A summary of the various SRM approaches. 12 SRM Proposed mechanism and Global mean Key climate and References approach associated uncertainties of negative radiative environmental the SRM approach forcing potential effects and characteristics Stratospheric Injection of aerosols or their 1–8 W m-2, depending on Change in temperature (Visioni et al., Aerosol precursor gases into the stratosphere the amount and pattern of and precipitation 2017; Tilmes et to scatter sunlight back to space; injection, and transport pattern; precipitation al., 2018b; Injection Aerosol types such as sulphates, and growth of injected reduction in some Simpson et al., (SAI) calcium carbonate, and titanium particles; compared to monsoon regions; 2019a) dioxide have been proposed; large other SRM approaches, decrease in direct and uncertainties associated with type of radiative forcing could increase in diffuse aerosol, aerosol radiative properties, be more homogenously sunlight at surface; microphysics, chemistry, distributed. stratospheric heating stratospheric processes, and and changes to temporal and spatial strategy of stratospheric dynamics aerosol injection and chemistry; potential delay in ozone hole recovery; changes in surface UV radiation; changes in crop yields Marine cloud Injection of sea salt or other types of 1–5 W m-2, depending on Change in land-sea (Latham et al., brightening aerosols to increase the albedo of the scale and amount of contrast and 2012)(Latham et marine stratocumulus clouds; sea salt injection; precipitation patterns al., 2014) (Ahlm (MCB) regional option to reduce SST in heterogeneous radiative et al., 2017) hurricane formation regions and in forcing (Stjern et al., coral reef areas; large uncertainties 2018) associated with cloud microphysics and aerosol–cloud-radiation interactions. Cirrus cloud Inject ice nuclei in the upper 1-2 W m-2, depending on Changes in temperature (Storelvmo and thinning troposphere to reduce the lifetime cirrus microphysical and precipitation Herger, 2014; and optical thickness of cirrus clouds response and seeding pattern; increase in solar Jackson et al., (CCT) to allow more longwave radiation to strategy; heterogeneous radiation reaching 2016; Gasparini et escape to space; large uncertainties radiative forcing; loss in surface al., 2020) associated with cirrus cloud cirrus clouds could also formation processes, cirrus cause significant microphysics, and interaction with shortwave forcing aerosol regionally; risk of overseeding and consequent warming Surface-based Increase ocean albedo by creating Radiative forcing of a Change in land-sea (Evans et al., albedo microbubbles; add reflective few W m-2 might be contrast and 2010; Davin et material to incease desert albedo; achieved via increase in precipitation pattern for al., 2014; Zhang modification paint the roof of buildings white to ocean and desert albedo, ocean and desert albedo et al., 2016; Field increase roof reflectivity; increase but the large-scale increase; more localized et al., 2018; albedo of agriculture land via no-till implementation is not effect for white roofs, Kravitz et al., farming or modifying crop albedo, feasible; less than 0.5 W no-till farming, and crop 2018) add reflective material to increase m-2 for white roof and albedo modification. sea ice albedo crop albedo enhancement; heterogeneous radiative forcing. Do Not Cite, Quote or Distribute 4-84 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 [END TABLE 4.7 HERE] 3 4 5 AR5 assessed the climate response to, as well as risks and side effects of, several SRM options (Boucher et 6 al., 2013) and concluded with high confidennce that SRM, if practicable, could substantially offset a global 7 temperature rise and partially offset some other impacts of global warming, but the compensation for the 8 climate change caused by GHGs would be imprecise. AR5 furthermore concluded that models consistently 9 suggest that SRM would generally reduce climate differences compared to a world with elevated GHG 10 concentrations and no SRM; however, there would also be residual regional differences in climate (e.g., tem- 11 perature and rainfall) when compared to a climate without elevated GHGs. AR5 concluded with high 12 confidence that scaling SRM to substantial levels would carry the risk that if the SRM were terminated for 13 any reason, surface temperatures would increase rapidly (within a decade or two) to values consistent with 14 the GHG forcing (Boucher et al., 2013). 15 16 SR1.5 (de Coninck et al., 2018) assessed SRM in terms of its potential to limit warming to below 1.5°C in 17 temporary overshoot scenarios and the associated impacts. It concluded that SAI could limit warming to 18 below 1.5°C but that the climate response to SAI is uncertain and varies across climate models. Overall, the 19 assessment concluded that the combined uncertainties related to SRM approaches, including technological 20 maturity, limited physical understanding of the response to SRM, potential impacts, and challenges of 21 governance, constrain potential deployment of SRM in the near future. 22 23 This subsection assesses the global and large-scale physical climate system response to SRM based on 24 theoretical and modelling studies. There is no mature technology today to implement any of the SRM 25 options assessed here. A short summary of the SRM options, including the proposed mechanism of each 26 SRM approach, radiative forcing potential, and key climate and environmental effects, is listed in Table 4.7 27 Chapter 5 (Section 5.6.3) assesses the biogeochemical implications of SRM, Chapter 6 (Section 6.3.6) 28 assesses the potential ERF of the aerosol-based SRM options and Chapter 8 (Section 8.6.3) assesses the 29 abrupt water cycle changes in response to initiation or termination of SRM. The risks to human and natural 30 systems, impacts of SRM, ethics, and perceptions are assessed in the WGII report (Chapter 16). Governance 31 issues associated with SRM research and deployment are assessed in the WGII and WGIII Reports. The 32 assessment of technical feasibility and engineering aspects of SRM is beyond the scope of this report. 33 34 The AR5 assessed SRM modelling mainly based on idealized simulations that used solar constant 35 reductions. Since then, more in-depth investigations into specific SRM approaches have been conducted with 36 more sophisticated treatment of aerosol–cloud–radiative interactions and stratospheric dynamics and 37 chemistry underlying SAI, MCB, and CCT. Another major development since AR5 is the investigation into 38 whether multiple climate policy goals may be met by optimally designed SRM strategies, including large- 39 ensemble SAI simulations using multiple injection locations. There are large uncertainties in important 40 SRM-related processes such as aerosol microphysics and aerosol-cloud-radiation interaction and hence the 41 level of understanding is low. 42 43 As assessed in SR1.5 (de Coninck et al., 2018), most of the knowledge about SRM is based on idealized 44 model simulations and some natural analogues. In addition to single-model studies, more results from the 45 coordinated modelling work of Geoengineering Model Intercomparison Project (GeoMIP) have become 46 available. GeoMIP was initiated at the time of AR5 (Kravitz et al., 2011) (Kravitz et al., 2013a) and is now 47 in its second phase under the framework of CMIP6 (GEOMIP6, Kravitz et al., 2015). However, studies 48 based on GeoMIP6 data are currently limited and hence the assessment on climate response to SRM here is 49 derived mostly from GeoMIP literature together with studies with single models. 50 51 Simple calculations and climate modelling studies show that about 2% extra solar irradiance reflected away 52 from Earth or a 1 percentage point increase in planetary albedo (0.31 to 0.32) would suffice to offset global 53 mean warming from a doubling of the CO2 concentration (The Royal Society, 2009)(Kravitz et al., 2013a) 54 (Kravitz et al., 2021). To offset the same amount of CO2-induced GSAT increase, different levels of ERF are Do Not Cite, Quote or Distribute 4-85 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 required for different methods of SRM (Schmidt et al., 2012; Russotto and Ackerman, 2018)(Modak et al., 2 2016)(Chiodo and Polvani, 2016)(Duan et al., 2018) (Krishnamohan et al., 2019)(Zhao et al., 2020). 3 4 As assessed in AR5 (Boucher et al., 2013), abruptly introducing SRM to fully offset global warming reduces 5 temperature toward 1850–1900 values with an e-folding time of only about 5 years (Matthews and Caldeira, 6 2007). A more realistic approach would be a slow ramp-up of SRM to offset further warming (MacCracken, 7 2016) (Tilmes et al., 2016). Modelling studies have consistently shown that SRM has the potential to offset 8 some effects of increasing GHGs on global and regional climate, including the melting of Arctic sea ice 9 (Moore et al., 2014) (Berdahl et al., 2014) and mountain glaciers (Zhao et al., 2017), weakening of Atlantic 10 meridional overturning circulation (AMOC) (Cao et al., 2016; Hong et al., 2017) (Tilmes et al., 2020), 11 changes in extremes of temperature and precipitation (Curry et al., 2014) (Ji et al., 2018)(Muthyala et al., 12 2018), and changes in frequency and intensity of tropical cyclone (Moore et al., 2015) (Jones et al., 2017). 13 14 The climate response to SRM depends greatly on the characteristics of SRM implementation approaches. 15 There could be substantial residual or overcompensating climate change at both the global and regional 16 scales and seasonal timescales (Irvine et al., 2016) (Kravitz et al., 2014) (Fasullo et al., 2018) (McCusker et 17 al., 2015) (Gertler et al., 2020) (Jiang et al., 2019). This is because the climate response to SRM options is 18 different from the response to GHG increase (Figure 4.38). For instance, when global mean warming is 19 offset by a uniform reduction in incoming sunlight, there is residual warming in the high latitudes and 20 overcooling in the tropics (Kravitz et al., 2013a; Kalidindi et al., 2015), and a reduction in tropical mean 21 rainfall (Tilmes et al., 2013). In simulations of stratospheric SO2 injection, SRM diminishes the amplitude of 22 the seasonal cycle of temperature at many high‐latitude locations, with warmer winters and cooler summers 23 (Jiang et al., 2019). Further, the rates of response could differ between surface temperature and slow 24 components in the climate system such as sea-level rise (Irvine et al., 2012; Jones et al., 2018). SRM 25 implemented at a moderate intensity, for example by offsetting half of the global warming, has the potential 26 to reduce negative effects such as reduced precipitation that are associated with fully offsetting global mean 27 warming (Irvine et al., 2019) (Irvine and Keith, 2020). 28 29 For the same amount of global mean cooling achieved, the pattern of climate response would depend on 30 SRM characteristics (Niemeier et al., 2013)(Muri et al., 2018)(Duan et al., 2018). This is illustrated in Figure 31 4.38 for temperature and precipitation change relative to a high-CO2 world for scenarios of CO2 reduction, 32 solar irradiance reduction, SAI, and MCB. The pattern differences for different methods are much larger for 33 precipitation than for temperature. The pattern of climate change resulting from SRM is also different from 34 that resulting from CO2 reduction (Figure 4.38). It is virtually certain that SRM approaches would not be 35 able to precisely offset the GHG-induced anthropogenic climate change at global and regional scales. 36 37 Because of different sensitivity of precipitation change to CO2 and solar forcings (Myhre et al., 2017), if 38 shortwave-based SRM is used to fully offset GHG-induced global mean warming, there would be a 39 overcompensation of GHG-induced increase in global mean precipitation (Kravitz et al., 2013a; Tilmes et 40 al., 2013; Irvine et al., 2016). Further, regional SRM approaches such as aerosol injections into the Arctic 41 stratosphere are likely to remotely influence on tropical monsoon precipitation by shifting the mean position 42 of ITCZ (Nalam et al., 2018). However, the shift could be avoided by simultaneously cooling the southern 43 hemisphere (MacCracken et al., 2013; Kravitz et al., 2016; Nalam et al., 2018). The SRM response of 44 precipitation minus evapotranspiration (P–E) is found to be smaller than that of precipitation because of 45 reduction in both precipitation and evapotranspiration (Tilmes et al., 2013; Nalam et al., 2018; Irvine et al., 46 2019). Thus, global mean soil moisture could be effectively maintained, though with significant regional 47 variability (Cheng et al., 2019). 48 49 The Geoengineering Large Ensemble Project (GLENS) has investigated achieving multiple climate policy 50 goals by adjusting the rate of stratospheric SO2 injection at four different latitudes. GSAT, the inter- 51 hemispheric temperature difference, and the equator-to-pole temperature gradient could be maintained 52 simultaneously at the year-2020 level under RCP 8.5 (Tilmes et al., 2018a). The possibility of using SAI to 53 simultaneously stabilize non-temperature metrics such as tropical precipitation and Arctic sea-ice extent is 54 also explored (Lee et al., 2020). Furthermore, the potential of achieving multiple climate policy goals by 55 combining two SRM approaches is also examined in a few modelling studies, with low confidence in the Do Not Cite, Quote or Distribute 4-86 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 outcome of combining various approaches and the related climate response (Boucher et al., 2017; Cao et al., 2 2017). 3 4 5 [START FIGURE 4.38 HERE] 6 7 Figure 4.38: Multi-model response per degree global mean cooling in temperature and precipitation in response 8 to CO2 forcing and SRM forcing. Top row shows the response to a CO2 decrease, calculated as the 9 difference between pre-industrial control simulation and abrupt 4 • CO2 simulations where the CO2 10 concentration is quadrupled abruptly from the pre-industrial level (11-model average); second row shows 11 the response to a globally uniform solar reduction, calculated as the difference between GeoMIP 12 experiment G1 and abrupt 4 • CO2 (11-model average); third row shows the response to stratospheric 13 sulphate aerosol injection, calculated as the difference between GeoMIP experiment G4 (a continuous 14 injection of 5Tg SO2 per year at one point on the equator into the lower stratosphere against the RCP4.5 15 background scenario) and RCP4.5 (6-model average); and bottom row shows the response to marine 16 cloud brightening, calculated as the difference between GeoMIP experiment G4cdnc (increase cloud 17 droplet concentration number in marine low cloud by 50% over the global ocean against RCP4.5 18 background scenario) and RCP4.5 (8-model average). All differences (average of years 11–50 of 19 simulation) are normalized by the global mean cooling in each scenario, averaged over years 11–50. 20 Diagonal lines represent regions where fewer than 80% of the models agree on the sign of change. The 21 values of correlation represent the spatial correlation of each SRM-induced temperature and precipitation 22 change pattern with the pattern of change caused by a reduction of atmospheric CO2. RMS (root mean 23 square) is calculated based on the fields shown in the maps (normalized by global mean cooling). Further 24 details on data sources and processing are available in the chapter data table (Table 4.SM.1). 25 26 [END FIGURE 4.38 HERE] 27 28 29 Stratospheric aerosol injection (SAI) 30 Most research has focused on SIA, the injection of sulphate particles or its precursor gases such as SO2, 31 which would then be oxidized to H2SO4. Injection of other types of aerosol particles, such as calcite 32 (CaCO3), titanium dioxide (TiO2), aluminium oxide (Al2O3), and engineered nanoparticles has also been 33 proposed (Keith, 2010)(Ferraro et al., 2011)(Pope et al., 2012)(Keith et al., 2016)(Jones et al., 34 2016a)(Weisenstein et al., 2015), but are much less studied compared to sulphate injection. The natural 35 analogue for sulphate aerosol injection is major volcanic eruptions (see Cross-Chapter Box 4.1), While 36 volcanic eruptions are not perfect analogues for SAI (Robock et al., 2013; Plazzotta et al., 2018a; Duan et al., 37 2019), studies on climate impacts of past volcanic eruptions can inform on the potential impact of 38 stratospheric sulphate injection. For example, emergent constraints (see Chapter 1 and 5) that relate the 39 climate system response to volcanic eruptions can be used to reduce uncertainty of the land surface 40 temperature response to SAI (Plazzotta et al., 2018b). 41 42 The cooling potential of SAI using sulphate aerosols depends on many factors (Visioni et al., 2017) 43 including the amount of injection (Niemeier and Timmreck, 2015), aerosol microphysics (Krishnamohan et 44 al., 2020), the spatial and temporal pattern of injection (Tilmes et al., 2017), response of stratospheric 45 dynamics and chemistry (Richter Jadwiga et al., 2018), and aerosol effect on cirrus clouds (Visioni et al., 46 2018). A negative radiative forcing of a few W m-2 (ranging from 1 to 8 W m-2) could be achieved depending 47 on the amount and location of SO2 injected into the stratosphere (Pitari et al., 2014)(Aquila et al., 48 2014)(Niemeier and Timmreck, 2015)(Kleinschmitt et al., 2018a)(Kleinschmitt et al., 2018a)(Kravitz et al., 49 2017)(Tilmes et al., 2018a). The simulated efficacy of SAI by emission of SO2 (radiative forcing per mass of 50 injection rate) generally decreases with the increase in injection rate because of the growth of larger particles 51 (about 0.5 microns) through condensation and coagulation reducing the mass scattering efficiency (Niemeier 52 and Timmreck, 2015; Kleinschmitt et al., 2018b). However, efficacy changes little for total injection rate up 53 to about 25 Tg Syr-1 when SO2 is injected at multiple locations simultaneously (Kravitz et al., 2017)(Tilmes 54 et al., 2018a). Differences in model representation of aerosol microphysics, evolution of particle size, 55 stratospheric dynamics and chemistry, and aerosol microphysics-radiation-circulation interactions all 56 contribute to the uncertainty in simulated cooling efficiency of SAI. Compared to sulphate aerosols, injection 57 of non-sulphate particles would result in different cooling efficacy, but understanding is limited (Weisenstein Do Not Cite, Quote or Distribute 4-87 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 et al., 2015)(Pope et al., 2012)(Jones et al., 2016a). 2 3 Earlier modelling studies focused on the effect of equatorial sulphate injection that tends to overcool the 4 tropics and undercool the poles. Compared to equatorial injection, off-equatorial injection at multiple 5 locations shows a closer resemblance to the baseline climate in many aspects, including temperature, 6 precipitation, and sea ice coverage (Kravitz et al., 2019). However, significant regional and seasonal residual 7 and overcompensating climate change is reported, including regional shifts in precipitation, continued 8 warming of polar oceans, and shifts in the seasonal cycle of snow depth and sea ice cover (Simpson et al., 9 2019a)(Fasullo et al., 2018)(Jiang et al., 2019). By appropriately adjusting the amount, latitude, altitude, and 10 timing of the aerosol injection, modelling studies suggest that SAI is conceptually able to achieve some 11 desired combination of radiative forcing and climate response (medium confidence) (Dai et al., 12 2018)(MacMartin et al., 2017)(Visioni et al., 2020b)(Lee et al., 2020). 13 14 There is large uncertainty in the stratospheric response to SAI, and the change in stratospheric dynamics and 15 chemistry would depend on the amount, size, type, location, and timing of injection. There is high 16 confidence that aerosol-induced stratospheric heating will play an important role in surface climate change 17 (Simpson et al., 2019a) by altering the effective radiative forcing (Krishnamohan et al., 2019), lower 18 stratosphere stability (Ferraro and Griffiths, 2016), quasi-biennial oscillation (QBO) (Aquila et al., 2014) 19 (Niemeier and Schmidt, 2017)(Kleinschmitt et al., 2018a), polar vortexes (Visioni et al., 2020a), and North 20 Atlantic Oscillation (Jones et al., 2021). Model simulations indicate stronger polar jets and weaker storm 21 tracks and a poleward shift of the tropospheric mid-latitude jets in response to stratospheric sulphate 22 injections in the tropics (Ferraro et al., 2015)(Richter Jadwiga et al., 2018), as the meridional temperature 23 gradient is increased in the lower stratosphere by the aerosol-induced heating. The aerosol-induced warming 24 would also offset some of the GHG-induced stratospheric cooling. Compared to equatorial injection, off- 25 equatorial injection is likely to result in reduced change in stratospheric heating, circulation, and QBO 26 (Richter Jadwiga et al., 2018)(Kravitz et al., 2019). Stratospheric ozone response to sulphate injection is 27 uncertain depending on the amount, altitude, and location of injection (WMO, 2018). It is likely that sulphate 28 injection would cause a reduction in polar column ozone concentration and delay the recovery of Antarctic 29 ozone hole (Pitari et al., 2014)(Richter Jadwiga et al., 2018)(Tilmes et al., 2018b), which would have 30 implications for UV radiation and surface ozone (Pitari et al., 2014)(Richter Jadwiga et al., 2018)(Tilmes et 31 al., 2018b)(Xia et al., 2017). Injection of non-sulphate aerosols is likely to result in less stratospheric heating 32 and ozone loss (Keith et al., 2016)(Weisenstein et al., 2015)(Pope et al., 2012). One side effect of SAI is 33 increased sulphate deposition at surface. A recent modelling study indicates that to maintain global 34 temperature at 2020 levels under RCP 8.5, increased sulphate deposition from stratospheric sulphate 35 injection could be globally balanced by the projected decrease in tropospheric anthropogenic SO2 emission, 36 but the spatial distribution of sulphate deposition would move from low to high latitudes (Visioni et al., 37 2020c). 38 39 Marine cloud brightening (MCB) 40 MCB involves injecting small aerosols such as sea salt into the base of marine stratocumulus clouds where 41 the aerosols act as cloud condensation nuclei (CCN). In the absence of other changes, an increase in CCN 42 would produce higher cloud droplet number concentration with reduced droplet sizes, increasing cloud 43 albedo. Increased droplet concentration may also increase cloud water content and optical thickness, but 44 recent studies suggest that liquid water path response to anthropogenic aerosols is weak due to the competing 45 effects of suppressed precipitation and enhanced cloud water evaporation (Toll et al., 2019). An analogue for 46 MCB are reflective, persistent ‘ship tracks’ observed after the passage of a sea-going vessel emitting 47 combustion aerosols into susceptible clouds (Chen et al., 2012) (Christensen and Stephens, 2011) 48 (Gryspeerdt et al., 2019). A recent study (Diamond et al., 2020) found a substantial increase in cloud 49 reflectivity from shipping in southeast Atlantic basin, suggesting that a regional-scale test of MCB in 50 stratocumulus‐dominated regions could be successful. 51 52 Modelling studies suggest that MCB has the potential to achieve a negative forcing of about 1 to 5 W m-2, 53 depending on the deployment area and strategies of cloud seeding (Partanen et al., 2012) (Stjern et al., 54 2018)(Alterskjær et al., 2013) (Ahlm et al., 2017) (Hill and Ming, 2012). Regional applications of MCB has 55 also been suggested for offsetting severe impacts from tropical cyclones whose genesis is associated with Do Not Cite, Quote or Distribute 4-88 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 higher SST (MacCracken, 2016) (Latham et al., 2014) and for protecting coral reefs from higher SST 2 (Latham et al., 2013). However, such regional approaches also involve large uncertainties in the magnitude 3 of the responses and consequences. 4 5 Several modelling studies suggest that the direct scattering effect by injected particles might also play an 6 important role in the cooling effect of MCB, but the relative contribution of aerosol–cloud and aerosol– 7 cloud–radiation effect is uncertain (Partanen et al., 2012) (Ahlm et al., 2017) (Kravitz et al., 2013b). Relative 8 to the high-GHG climate, it is likely that MCB would increase precipitation over tropical land due to the 9 inhomogeneous forcing pattern of MCB over ocean and land (medium confidence) (Alterskjær et al., 2013) 10 (Ahlm et al., 2017)(Stjern et al., 2018) (Niemeier et al., 2013) (Muri et al., 2018) (Bala et al., 2011). Because 11 of the high level of uncertainty associated with cloud microphysics and aerosol–cloud–radiation interaction 12 (See also Chapter 7, Section 7.3), the climate response to MCB is as uncertain. Results from global climate 13 models are subject to large uncertainty because of different treatment of cloud microphysics and inadequate 14 representation of sub-grid aerosol and cloud processes (Stjern et al., 2018) (Stuart et al., 2013) (Alterskjær 15 and Kristjánsson, 2013) (Connolly et al., 2014). Sea salt deposition over land (Muri et al., 2015) and the 16 effect of sea salt emission on atmospheric chemistry (Horowitz et al., 2020) are some of the potential side 17 effects of MCB. 18 19 Cirrus Cloud Thinning (CCT) 20 Cirrus clouds trap more outgoing thermal radiation than they reflect incoming solar radiation and thus have 21 an overall warming effect on the climate system (Mitchell and Finnegan, 2009). The aim of CCT is to 22 reduce cirrus cloud optical depth by increasing the heterogeneous nucleation via seeding cirrus clouds with 23 an optimal concentration of ice nucleating particles,which might cause larger ice crystals and rapid fallout, 24 resulting in reduced lifetime and coverage of cirrus clouds (Muri et al., 2014), (Gasparini et al., 2017; 25 Lohmann and Gasparini, 2017; Gruber et al., 2019a). CCT aims to achieve the opposite effect of contrails 26 that increase cirrus cover and cause a small positive ERF (Chapter 7, Section 7.3). A high-resolution 27 modelling study of CCT over a limited area of the Arctic suggested that cirrus seeding causes a decrease in 28 ice crystal number concentration and a reduction in mixed-phase cloud cover, both of which cause a cooling 29 effect (Gruber et al., 2019b). 30 31 Under present-day climate, cirrus clouds exerts a net positive radiative forcing of about 5 W m-2 (Gasparini 32 and Lohmann, 2016)(Hong et al., 2016), indicating a maximum cooling potential of the same magnitude if 33 all cirrus cloud were removed from the climate system. However, modelling results show a much smaller 34 cooling effect of CCT. For the optimal ice nuclei seeding concentration and globally non-uniform seeding 35 strategy, a net negative cloud radiative forcing of about 1 to 2 W m-2 is achieved (Storelvmo and Herger, 36 2014) (Gasparini et al., 2020). A few studies find that no seeding strategy could achieve a significant cooling 37 effect, owing to complex microphysical mechanisms limiting robust climate responses to cirrus seeding 38 (Penner et al., 2015; Gasparini and Lohmann, 2016). A higher than optimal concentration of ice nucleating 39 particles could also result in over-seeding that increases rather than decreases cirrus optical thickness 40 (Storelvmo et al., 2013) (Gasparini and Lohmann, 2016). Thus, there is low confidence in the cooling effect 41 of CCT, due to limited understanding of cirrus microphysics, its interaction with aerosols, and the 42 complexity of seeding strategy. 43 44 Relative to the high-GHG climate and for the same amount of global cooling, CCT is simulated to cause an 45 increase in global precipitation compared to shortwave-based SRM options such as SAI and MCB (Muri et 46 al., 2018) (Duan et al., 2018) because of the opposing effects of CCT and increased CO2 on outgoing 47 longwave radiation (Kristjánsson et al., 2015) (Jackson et al., 2016). Combining SAI and CCT has suggested 48 that GHG-induced changes in global mean temperature and precipitation can be simultaneously offset (Cao 49 et al., 2017), but there is low confidence in the applicability of this result to the real world owing to the large 50 uncertainty in simulating aerosol forcing and the complex cirrus microphysical processes. 51 52 Surface-based albedo modification 53 Surface-based albedo modification could, in principle, achieve a negative radiative forcing of a few W m-2 54 by enhancing the albedo of the ocean surface (Kravitz et al., 2018)(Gabriel et al., 2017). However, the 55 technology does not exist today to increase ocean albedo at large scale. An increase in crop albedo or roof Do Not Cite, Quote or Distribute 4-89 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 albedo in urban areas could help to reduce warming in densely populated and important agricultural regions, 2 but the effect would be limited to local scales and ineffective at counteracting global warming (Crook et al., 3 2015a) (Zhang et al., 2016). Large changes in desert albedo could in principle result in substantial global 4 cooling, but would severely alter the hydrological cycle (Crook et al., 2015a). 5 6 In addition to above-mentioned SRM methods, a number of local intervention methods have been proposed 7 to limit the loss of cryosphere, such as applying reflective materials over sea ice (Field et al., 2018), pumping 8 seawater on top of the ice surface (Desch et al., 2017) (Zampieri and Goessling, 2019), depositing massive 9 amount of snow over ice sheets (Feldmann et al., 2019), and blocking warm seawater from reaching glaciers 10 (Moore et al., 2018a). The stabilization of ice sheets through local intervention methods would reduce sea 11 level commitment (Section 9.6.3.5). However, these methods are subject to large uncertainty concerning 12 their feasibility and effectiveness, and their effects would be largely localized. 13 14 Detectability of climate response to SRM 15 Internal variability could mask the response to SRM-related forcing in the near term (see also Section 16 4.6.3.1). A detection of the global scale climate system response to stratospheric sulphate aerosol injection 17 will likely require a forcing of the size produced by the 1991 Mount Pinatubo eruption (Robock et al., 2010). 18 In model simulations of where 5 Tg SO2 is injected into the stratosphere continuously (roughly one fourth of 19 the 1991 Pinatubo eruption per year) under RCP 4.5, it is shown that, relative to the high-GHG world 20 without SRM, the effect of SRM on global temperature and precipitation is detectable after one to two 21 decades (Bürger and Cubasch, 2015; Lo et al., 2016) which is similar to the timescale for the emergence of 22 GSAT trends due to strong mitigation (Section 4.6.3.1). The detection time is sensitive to detection methods 23 and filtering techniques (Lo et al. 2016). An analysis using GLENS simulation (MacMartin et al., 2019) 24 compares response in temperature, precipitation, and precipitation minus evapotranspiration (P-E) between a 25 climate state with GHG-induced 1.5° C global mean temperature change and that with the same global mean 26 temperature but under RCP4.5 emissions and a limited deployment of SO2 injection. It is found that at grid- 27 scale, difference in climate response between these two climate states are not detectable by the end of this 28 century. However, for higher emission scenarios of the RCP8.5 and correspondingly larger SRM deployment 29 for maintaining the same global mean temperature change of 1.5°C, the regional differences are detectable 30 before the end of the century. In addition to surface temperature and precipitation, observations of aerosol 31 burden and temperature in the stratosphere via the deployment of stratospheric aerosol observing system 32 might facilitate the detection of climate response to SAI. 33 34 Climate response to termination of SRM 35 A hypothetical, sudden and sustained termination of SRM in a world with high GHG concentrations has 36 been simulated to cause climate rebound effects such as rapid increase in global temperature, precipitation, 37 and sea level, and rapid reduction in sea-ice area (Crook et al., 2015; Jones et al., 2013; McCusker et al., 38 2014; Muri et al., 2018). Model simulations also show reduced precipitation over land areas in the first few 39 years following termination, indicating general drying that would exacerbate the effects of rapid warming 40 (McCusker et al., 2014). A sudden and sustained termination of SRM is also expected to weaken carbon 41 sinks, accelerating atmospheric CO2 accumulation and warming (Tjiputra et al., 2016) (Muri et al., 2018) 42 (Plazzotta et al., 2019). A gradual phase-out of SRM combined with mitigation and CDR could reduce the 43 large warming rates from sudden SRM termination (MacMartin et al., 2014) (Keith and MacMartin, 2015; 44 Tilmes et al., 2016), though this would be limited by how rapidly emission reductions can be scaled up 45 (Ekholm and Korhonen, 2016). 46 47 Synthesis of the climate response to SRM 48 49 Modelling studies have consistently shown that SRM has the potential to offset some effect of increasing 50 GHGs on global and regional climate (high confidence), but there would be substantial residual or 51 overcompensating climate change at the regional scale and seasonal timescale (high confidence). Large 52 uncertainties associated with aerosol–cloud–radiation interactions persist in our understanding of climate 53 response to aerosol-based SRM options. For the same amount of global mean cooling, different SRM options 54 would cause different patterns of climate change (medium confidence). Modelling studies suggest that it is 55 conceptually possible to achieve multiple climate policy goals by optimally designed SRM strategies. Do Not Cite, Quote or Distribute 4-90 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 The effect of SRM options on global temperature and precipitation response would detectable after one or 3 two decades, which is similar to the timescale for the detection of strong mitigation. There is high confidence 4 that a sudden and sustained termination of a high level of SRM against a high-GHG background would cause 5 a rapid increase in temperature at a rate that far exceeds that projected for climate change without SRM. 6 However, a gradual phase-out of SRM combined with mitigation and CDR would more likely than not avoid 7 large rates of warming. 8 9 10 4.7 Climate Change Beyond 2100 11 12 This section assesses changes in climate beyond 2100. An advance since AR5 is the availability of ESM 13 results for scenarios beyond 2100 and for much longer stabilisation simulations compared with analysis 14 predominantly based on Earth system models of intermediate complexity (EMICs) at the time of AR5 (e.g. 15 Eby et al., 2013; Zickfeld et al., 2013). Long-term commitment of sea-level rise due to thermal expansion 16 and ice-sheet loss is assessed in Chapter 9 (9.6.3.5; figure 9.0). Here we assess projections of GSAT, global 17 precipitation, and Arctic sea ice. Uncertainties relating to potential long-term changes in AMOC are treated 18 in 9.2.3.1. 19 20 On multi-century timescales it is common to explore changes that are due to long-term commitment.Here we 21 differentiate between: 22 • Committed emissions due to infrastructure. Infrastructure that causes greenhouse gas emissions 23 cannot be changed straight away leading to a commitment from existing infrastructure that some 24 emissions will continue for a number of years into the future (Davis and Socolow, 2014; Smith et al., 25 2019a). Further consideration of this aspect of commitment will be assessed by WGIII. 26 • Climate response to constant emissions. Some of the scenario extensions beyond 2100 make 27 assumptions about constant emissions (either positive or negative). Section 4.7.1 will assess changes 28 in climate under scenario extensions beyond 2100. 29 • Committed climate change to constant atmospheric composition. There is widespread literature on 30 how the climate continues to change after stabilisation of radiative forcing. This includes diagnosing 31 the long-term climate response to a doubling of CO2 (ECS, Chapter 7). Since AR5, more GCMs 32 have run stabilised forcing simulations for many centuries allowing new insights into their very 33 long-term behaviour (Section 7.4.3). 34 • Committed response to zero emissions. How climate would continue to evolve if all emissions 35 ceased. SR1.5 assessed changes in climate if emissions of all greenhouse gases and aerosols ceased. 36 Section 4.7.2 assesses new results considering cessation of CO2-only emissions which forms a 37 significant term in calculating remaining carbon budgets. 38 • Irreversibility. Some changes do not revert if the forcing is removed, leaving a committed change to 39 the system. Section 4.7.2 assesses changes in the Earth system which may be irreversible. 40 • Abrupt changes. If a tipping point in the climate system is passed, then some elements may continue 41 to respond if the forcing which caused them is removed. Section 4.7.2 assesses the potential for 42 abrupt changes in the Earth system. 43 44 45 4.7.1 Commitment and Climate Change Beyond 2100 46 47 4.7.1.1 Climate change following zero emissions 48 49 The zero emissions commitment (ZEC) is the the climate change commitment that would result, in terms of 50 projected GSAT, from setting carbon dioxide (CO2) emissions to zero. It is determined by both inertia in 51 physical climate system components (ocean, cryosphere, land surface) and carbon cycle inertia (see Annex 52 VII). In its widest sense it refers to emissions of all compounds including greenhouses gases, aerosols and 53 their pre-cursors. A specific sub-category of zero emissions commitment is the zero CO2 emissions 54 commitment, which refers to the climate system response to a cessation of anthropogenic CO2 emissions Do Not Cite, Quote or Distribute 4-91 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 excluding the impact of non-CO2 forcers. Assessment of remaining carbon budgets requires an assessment of 2 zero CO2 emissions commitment as well as of the transient climate response to cumulative carbon emissions 3 (TCRE, Chapter 5 Section 5.5.2). 4 5 There is an offset of continued warming following cessation of emissions by continued CO2 removal by 6 natural sinks (e.g. Joos et al., 2013; Matthews and Caldeira, 2008; Solomon et al., 2009; Ricke and Caldeira, 7 2014) (high confidence). Some models continue warming by up to 0.5°C after emissions cease at 2°C of 8 warming (Frölicher & Paynter, 2015; Frölicher et al., 2014; Williams et al., 2017), while others simulate 9 little to no additional warming (Nohara et al., 2015). In SR1.5, the available evidence indicated that past CO2 10 emissions do not commit to substantial further warming (Allen et al., 2018). A ZEC close to zero was thus 11 applied for the computation of the remaining carbon budget for the (Rogelj et al., 2018b). However, the 12 available literature consisted of simulations from a small number of models using a variety of experimental 13 designs, with some simulations showing a complex evolution of temperature following cessation of 14 emissions (e.g., Frölicher & Paynter, 2015; Frölicher et al., 2014). 15 16 Here we draw on new simulations to provide an assessment of ZEC using multiple ESMs (Jones et al., 17 2019b) and EMICs (MacDougall et al., 2020). Figure 4.39 shows results from 20 models that simulate the 18 evolution of CO2 and the GSAT response following cessation of CO2 emissions for an experiment where 19 1000 PgC is emitted during a 1% per year CO2 increase. All simulations show a strong reduction in 20 atmospheric CO2 concentration following cessation of CO2 emissions in agreement with previous studies and 21 basic theory that natural carbon sinks will persist. Therefore, there is very high confidence that atmospheric 22 CO2 concentrations would decline for decades if CO2 emissions cease. Temperature evolution in the 100 23 years following cessation of emissions varies by model and across timescales, with some models showing 24 declining temperature, others having ZEC close to zero, and others showing continued warming following 25 cessation of emissions (Figure 4.39). The GSAT response depends on the balance of carbon sinks and ocean 26 heat uptake (MacDougall et al., 2020). The 20-year average GSAT change 50 years after the cessation of 27 emissions (ZEC50) is summarised in Table 4.8. The mean value of ZEC50 is –0.079°C, with 5–95% range – 28 0.34 to 0.28 °C. There is no strong relationship between ZEC50 and modelled climate sensitivity (neither 29 ECS nor TCR; MacDougall et al., 2020). It is therefore likely that the absolute magnitude of ZEC50 is less 30 than 0.3 °C, but we assess low confidence in the sign of ZEC on 50-year timescales. This is small compared 31 with natural variability in GSAT. 32 33 34 [START FIGURE 4.39 HERE] 35 36 Figure 4.39: Zero Emissions Commitment (ZEC). Changes in (a) atmospheric CO2 concentration and (b) evolution 37 of GSAT following cessation of CO2 emissions branched from the 1% per year experiment after 38 emission of 1000 PgC (Jones et al., 2019a). ZEC is the temperature anomaly relative to the estimated 39 temperature at the year of cessation. ZEC50 is the 20-year mean GSAT change centred on 50 years after 40 the time of cessation (see Table 4.8) – this period is marked with the vertical dotted lines. Multi-model 41 mean is shown as thick black line, individual model simulations are in grey. Further details on data 42 sources and processing are available in the chapter data table (Table 4.SM.1). 43 44 [END FIGURE 4.39 HERE] 45 46 47 [START TABLE 4.8 HERE] 48 49 Table 4.8: The 20-year average GSAT change 50 years after the cessation of emission (ZEC50). Displayed are 50 ZEC50 estimated from eleven ESMs (top) and nine EMICs (bottom). 51 MODEL ZEC50 (°C) ACCESS-ESM1.5 0.01 CANESM5 -0.14 CESM2 -0.31 CNRM-ESM2-1 0.06 GFDL-ESM2M -0.27 Do Not Cite, Quote or Distribute 4-92 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI GFDL-ESM4 -0.21 GISS-E2-1-G -0.15 MIROC-ES2L -0.08 MPI-ESM1.2-LR -0.27 NORESM2-LM -0.33 UKESM1-0-LL 0.28 BERN3D-LPX 0.01 DCESS1.0 0.06 CLIMBER-2 -0.07 IAPRAS 0.28 LOVECLIM 1.2 -0.04 MESM 0.01 MIROC-LITE -0.06 PLASM-GENIE -0.36 UVIC ESCM 2.10 0.03 1 2 [END TABLE 4.8 HERE] 3 4 5 4.7.1.2 Change in Global Climate Indices Beyond 2100 6 7 This subsection assesses changes in global climate indices out to 2300 using extensions of the SSP scenarios 8 (Meinshausen et al., 2020) and literature based on extensions to the RCP scenarios from CMIP5 9 (Meinshausen et al., 2011), which differ from the SSPs despite similar labelling of global radiative forcing 10 levels (Section 4.6.2). Meinshausen et al. (2020) describe the extensions to the SSP scenarios, which differ 11 slightly from the ScenarioMIP documentation (O’Neill et al., 2016). A simplified approach across scenarios 12 reduces emissions such that after 2100, land use CO2 emissions are reduced to zero by 2150; any net 13 negative fossil CO2 emissions are reduced to zero by 2200, and positive fossil CO2 emissions are reduced to 14 zero by 2250. Non-CO2 fossil fuel emissions are also reduced to zero by 2250 while land-use-related non- 15 CO2 emissions are held constant at 2100 levels. The extensions are created up to the year 2500, but ESM 16 simulations have only been requested, as part of the CMIP6 protocol, to run to 2300. As a result, unlike the 17 RCP8.5 extension, SSP5-8.5 sees a decline in CO2 concentration after 2250, but the radiative forcing level is 18 similar, reaching approximately 12 Wm-2 during most of the extension. Both SSP1-2.6 and SSP5-3.4-OS 19 decrease radiative forcing after 2100. SSP5-3.4-OS is designed to return to the same level of forcing as 20 SSP1-2.6 during the first half of the 22nd century. Because relatively few CMIP6 ESMs have submitted 21 results beyond 2100, GSAT projections using the MAGICC7 emulator (see Cross-Chapter Box 7.1) are also 22 shown here. 23 24 Changes in climate at 2300 have impacts and commitments beyond this timeframe (high confidence). Sea- 25 level rise may exceed 2 m on millennial timescales even when warming is limited to 1.5–2°C, and tens of 26 meters for higher warming levels (Chapter 9, Section 9.6.3.5, Table 9.10). Randerson et al. (2015) showed 27 increasing importance on carbon cycle feedbacks of slow ocean processes, Mahowald et al. (2017) showed 28 the long-lasting legacy of land-use effects and Moore et al., (2018) show how changes in Southern Ocean 29 winds affect nutrients and marine productivity well beyond 2300. Clark et al. (2016) show that physical and 30 biogeochemical impacts of 21st century emissions have a potential committed legacy of at least 10,000 31 years. 32 33 34 [START FIGURE 4.40 HERE] 35 36 Figure 4.40: Simulated climate changes up to 2300 under the extended SSP scenarios. Displayed are (a) projected 37 GSAT change, relative to 1850–1900, from CMIP6 models (individual lines) and MAGICC7 (shaded 38 plumes), (b) as (a) but zoomed in to show low-emission scenarios, (c) global land precipitation change, 39 and (d) September Arctic sea-ice area. Further details on data sources and processing are available in the 40 chapter data table (Table 4.SM.1). 41 42 [END FIGURE 4.40 HERE] 43 Do Not Cite, Quote or Distribute 4-93 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 4.7.1.2.1 Global Surface Air Temperature 3 Both CMIP6 and CMIP5 results show that global temperature beyond 2100 is strongly dependent on 4 scenario, and the difference in GSAT projections between high- and low-emission scenarios continues to 5 increase (high confidence). Under the extended RCP2.6 (Caesar et al., 2013) and SSP1-2.6 scenarios, where 6 CO2 concentration and radiative forcing continue to decline beyond 2100, GSAT stabilises during the 21st 7 century before decreasing and remaining below 2°C until 2300 except in some of the very high climate- 8 sensitivity ESMs, which project GSAT to stay above 2°C by 2300 (Figure 4.40). Under RCP8.5, regional 9 temperature changes above 20°C have been reported in multiple models over high-latitude land areas (Caesar 10 et al., 2013; Randerson et al., 2015b). Non-CO2 forcing and feedbacks remain important by 2300 (high 11 confidence). Randerson et al. (2015) found that 1.6°C of warming by 2300 came from non-CO2 forcing alone 12 in RCP8.5, and Rind et al. (2018) show that regional forcing from aerosols can have notable effects on ocean 13 circulation on centennial timescales. High latitude warming led to longer growing seasons and increased 14 vegetation growth in the CESM1 model (Liptak et al., 2017), and Burke et al. (2017) found that carbon 15 release from permafrost areas susceptible to this warming may amplify future climate change by up to 17% 16 by 2300. 17 18 Too few CMIP6 models performed the extension simulations to allow a robust assessment of GSAT 19 projection, and some of those which did had higher than average climate sensitivity values. Therefore, we 20 base our assessment of GSAT projections (Table 4.9) on the MAGICC7 emulator calibrated against assessed 21 GSAT to 2100 (Section 4.3.4, Cross-Chapter Box 7.1). Because the emulator approach has not been 22 evaluated in depth up to 2300 in the same way as it has up to 2100 (Cross-Chapter Box 7.1) we account for 23 possible additional uncertainty by assessing the 5–95% range from MAGICC as likely instead of very likely. 24 It is therefore likely that GSAT will exceed 2°C above that of the period 1850–1900 at the year 2300 in the 25 extended SSP scenarios SSP2-4.5, SSP3-7.0 and SSP5-8.5 (Figure 4.40). For SSP1-2.6 and SSP1-1.9, mean 26 warming at 2300 is 1.5°C and 0.9°C respectively. GSAT differences between SSP5-3.4-overshoot and SSP1- 27 2.6 peak during the 21st century but decline to less than about 0.25°C after 2150 (medium confidence). 28 29 To place the temperature projections for the end of the 23rd century into the context of paleo temperatures, 30 GSAT under SSP2-4.5 (2.3°C–4.6°C) has not been experienced since the Mid Pliocene, about 3 million 31 years ago. GSAT projected for the end of the 23rd century under SSP5-8.5 (6.6°C–14.1°C) overlaps with the 32 range estimated for the Miocene Climatic Optimum (5-10°C) and Early Eocene Climatic Optimum (10°C– 33 18°C), about 15 and 50 million years ago, respectively (medium confidence; Chapter 2). 34 35 36 [START TABLE 4.9 HERE] 37 38 Table 4.9: Change of global surface air temperature at 2300. Displayed are the median and 5–95% range of 39 GSAT change at 2300 relative to 1850–1900 for the six scenarios used with MAGICC7. Scenario Median (°C) 5–95% range (°C) SSP5-8.5 9.6 6.6–14.1 SSP3-7.0 8.2 5.7–11.8 SSP2-4.5 3.3 2.3–4.6 SSP5-3.4-OS 1.6 1.1–2.2 SSP1-2.6 1.5 1.0–2.2 SSP1-1.9 0.9 0.6– .4 40 41 [END TABLE 4.9 HERE] 42 43 44 4.7.1.2.2 Global Land Precipitation 45 Global land precipitation will continue to increase in line with GSAT under high emission scenarios 46 (medium confidence). Precipitation changes over land show larger variability and a less clear signal than 47 global total precipitation. Caesar et al. (2013) showed that under the CMIP5 extension simulations, 48 HadGEM2-ES projected global land precipitation to remain roughly the same in RCP2.6, to increase by Do Not Cite, Quote or Distribute 4-94 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 about 4% in RCP4.5 and to increase by about 7% in RCP8.5. Their results showed global precipitation 2 increasing linearly with temperature while radiative forcing increases, but then more quickly if forcing is 3 stabilised or reduced. This backs up findings of an intensification of the hydrological cycle following CO2 4 decrease which has been attributed to a build-up of ocean heat (Wu et al., 2010), and to a fast atmospheric 5 adjustment to CO2 radiative forcing (Cao et al., 2011a). Figure 4.40 shows that global land precipitation 6 increases in CMIP6 models until 2300 for SSP5-8.5 but stabilises in SSP1-2.6 and SSP5-3.4-OS. SSP1-2.6 7 and SSP5-3.4-OS are not distinguishable in behaviour of projected global land precipitation after 2100. 8 9 10 4.7.1.2.3 Arctic Sea Ice 11 Chapter 9 assesses with high confidence that on decadal and longer timescales, Arctic summer sea-ice area 12 will remain highly correlated with global mean temperature until the summer sea ice has vanished (Section 13 9.3.1.1). This means that Arctic sea ice will continue to decline in scenarios of continued warming but will 14 begin to recover in scenarios where GSAT begins to decrease. Under the CMIP5 extension simulations, 15 minimum (September) Arctic sea-ice area began to recover for most models under RCP2.6 out to 2300, 16 while RCP4.5 and RCP8.5 extensions became ice-free in September (Hezel et al., 2014; Bathiany et al., 17 2016). They also found increasingly strong winter responses under continued warming such that under the 18 RCP8.5 extension, the Arctic became ice-free nearly year-round by 2300. Consistent with the assessment in 19 Section 9.3.1.1 that Arctic sea-ice area is correlated with GSAT, CMIP6 projections to 2300 show partial sea 20 ice recovery by 2300 in SSP1-2.6 in line with GSAT (Figure 4.40), with one model (MRI-ESM2-0) showing 21 near complete recovery to present-day values. SSP1-2.6 and SSP5-3.4-OS are not distinguishable in 22 behaviour of Arctic sea-ice in these models after 2100. SSP5-8.5 remains ice-free in September up to 2300. 23 24 25 4.7.2 Potential for Abrupt and Irreversible Climate Change 26 27 Similar to AR5 and SROCC, AR6 defines an abrupt climate change as a large-scale abrupt change in 28 the climate system that takes place over a few decades or less, persists (or is anticipated to persist) for at least 29 a few decades and causes substantial impacts in human and/or natural systems (Annex VII: Glossary). 30 Further, AR6 considers such a perturbed state of a dynamical system as irreversible on a given timescale, if 31 the recovery timescale from this state due to natural processes takes substantially longer than the timescale of 32 interest (Annex VII: Glossary). The AR6 adopts the related definition of a tipping point as a critical 33 threshold beyond which a system reorganizes, often abruptly and/or irreversibly, and a tipping element as a 34 component of the Earth system that is susceptible to a tipping point (Annex VII: Glossary). Tipping points 35 may involve global or regional climate changes from one stable state to another stable state or to changes 36 that occur faster than the rate of change of forcing (Alley et al., 2003) and include shifts from one 37 equilibrium state to another and other responses of the climate system to external forcing (see Section 1.2.4.2 38 in Chapter 1). While reversibility has been defined alternatively in the literature with respect to the response 39 specifically to idealized CO2 forcing and generally GSAT change, AR6 considers both definitions 40 synonymous, because it has been widely demonstrated that the GSAT change is reversible in models with 41 respect to CO2 with a several-year lag (Boucher et al., 2012). 42 43 Abrupt and irreversible changes in the climate system are assessed across multiple chapters in AR6. This 44 section provides a cross-chapter synthesis of these assessments as an update to the AR5 Table 12.4 and 45 SROCC Table 6.1. Understanding of abrupt climate change and irreversibility has advanced considerably 46 since AR5 with many of the projected changes in proposed Tipping Elements having grown more confident 47 (Table 4.10). Many aspects of the physical climate changes induced by GHG warming previously 48 demonstrated to be reversible in a single model have been confirmed in multiple models (Boucher et al., 49 2012; Tokarska and Zickfeld, 2015) with others such as sea-level rise or terrestrial ecosystems confirmed to 50 continue to respond on long timescales (Clark et al., 2016; Zickfeld et al., 2017; Pugh et al., 2018). 51 52 The Carbon Dioxide Removal Model Intercomparison Project (CDR-MIP) (Keller et al., 2018) comprises a 53 set of 1% ramp-up ramp-down simulations aimed at establishing a multi-model assessment of reversibility of 54 Earth system components. Preliminary results from CDRMIP are presented in Section 4.6.3. Results from 55 the SSP5-3.4-Overshoot scenario and other quantities of climate change at the same CO2 level before and Do Not Cite, Quote or Distribute 4-95 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 after overshoot are assessed in Section 4.6.2. Forcing reversal is followed by reversal of ocean surface and 2 land temperature along with land and ocean precipitation, snow cover, and Arctic sea ice with a lag of a few 3 years to decades (Table 4.10). Other tipping elements have much longer timescales of reversibility from 4 decades to millennia. (Drijfhout et al., 2015) provided an assessment of 13 regional mechanisms of abrupt 5 change, finding abrupt changes in sea ice, oceanic flows, land ice, and terrestrial ecosystem response, 6 although with little consistency among the models. The potential for abrupt changes in ice sheets, the 7 AMOC, tropical forests, and ecosystem responses to ocean acidification were also recently reviewed by 8 (Good et al., 2018). They found that some degree of irreversible loss of the West Antarctic Ice Sheet (WAIS) 9 may have already begun, that tropical forests are adversely affected by drought, and rapid development of 10 aragonite undersaturation at high latitudes affecting calcifying organisms. 11 12 New since AR5 is the fundamental recognition in SRCCL and in this Report (Chapter 5) that projected 13 changes in forests strongly depend on the human disturbance and that tropical forest dieback in the absence 14 of disturbance is largely driven by the increased potential for drought, while that in boreal forests includes 15 both thermal and hydrological factors (Drijfhout et al., 2015). For some proposed tipping elements, the role 16 of seasonal change has become better understood. For example, the lack of a tipping point in the reduction 17 of summer Arctic sea-ice area (Stroeve and Notz, 2015) has been further substantiated. The role of abrupt 18 change at the edges (Bathiany et al., 2020) has also been clarified, as has been the importance of 19 distinguishing summer from winter mechanisms and associated abruptness, because ice area reduces 20 gradually in summer, but not necessarily in winter (Bathiany et al., 2016). For other tipping elements 21 including AMOC (Section 9.2.3.1), mixed layer depth (9.2.1.3), and sea-level rise (9.6.3.5), an increase in 22 the diversity of model structure and sensitivity to multiple factors has led to a better understanding of the 23 complexity of the problem, with some increase in assessed uncertainty and an assessed deep uncertainty (see 24 Annex VII: Glossary) related to projected sea-level rise with global warming levels above 3°C (Section 25 9.6.3.5). In still other cases such as Antarctic Sea Ice (Section 9.3.2) and Southern Ocean Meridional 26 Overturning Circulation (MOC; Section 9.2.3.1), uncertainty remains high. Finally, it has also been 27 postulated that models may be prone to being too stable (Valdes, 2011) based on the limitations of models as 28 well as other lines of evidence such paleo-evidence of abrupt events (Dakos et al., 2008; Klus et al., 2018; 29 Sime et al., 2019). 30 31 32 [START TABLE 4.10 HERE] 33 34 Table 4.10: Cross-chapter assessment updating AR5 and SROCC of components in the Earth system that have been 35 proposed as susceptible to tipping points/abrupt change, irreversibility, projected 21st century change, 36 and overall change in assessment from previous IPCC reports. Also provided are confidence levels and, 37 in parentheses, the main section(s) of this report in which proposed tipping elements are assessed. 38 Earth System Potential Abrupt Irreversibility if forcing Projected 21st century change under continued Change in Assessment Component/Tipping Climate Change? reversed (timescales warming Element indicated) Global Monsoon (4.5.1.5; Yes under Reversible within years to Medium confidence in global monsoon increase; More lines of evidence 8.6) AMOC collapse, decades, Medium confidence Medium confidence in Asian-African than AR5 medium strengthening and North American weakening confidence Tropical Forest (5.4.8; 8.6.2) Yes, Low Irreversible for multi- Medium confidence of increasing vegetation More confident rates confidence decades, Medium confidence carbon storage depending on human disturbance than AR5 Boreal Forest (5.4.8) Yes, Low Irreversible for multi- Medium confidence in offsetting lower latitude More confident rates confidence decades, Medium confidence dieback and poleward extension depending on than AR5 human disturbance Permafrost Carbon (5.4.8) Yes, High Irreversible for centuries, Virtually certain decline in frozen carbon; Low More confident rates confidence High confidence confidence in net carbon change than SROCC Do Not Cite, Quote or Distribute 4-96 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI Arctic Summer Sea Ice No, high Reversible within years to Likely complete loss More specificity than (4.3.2; 4.6.2.1; 9.3.1) confidence decades, High confidence SROCC Arctic Winter Sea Ice (4.3.2; Yes, High Reversible within years to High confidence in moderate winter declines More specificity than 9.3.1) confidence decades, High confidence SROCC Antarctic Sea Ice (9.3.2) Yes, Low Unknown, Low confidence Low confidence in moderate winter and summer Improved CMIP6 confidence declines simulation Greenland Ice Sheet (9.4.1) No, High Irreversible for millennia, Virtually certain mass loss under all scenarios More lines of evidence confidence High confidence than SROCC West Antarctic Ice Sheet Yes, High Irreversible for decades to Likely mass loss under all scenarios; Deep Added deep uncertainty and Shelves (9.4.2; Box 9.4) confidence millennia, High confidence uncertainty in projections for above 3°C at GWL > 3°C Global Ocean Heat Content No, High Irreversible for centuries, Very high confidence oceans will continue to Better consistency with (4.5.2.1; 4.6.2.1; 9.2.2; confidence Very high confidence warm ECS/TCR CCBox 7.1) Global Sea-Level Rise Yes, High Irreversible for centuries, Very high confidence in continued rise; Deep Added deep uncertainty (4.6.2.1; 4.6.3.2; 9.6.3.5; confidence Very high confidence uncertainty in projections for above 3°C at GWL > 3°C Box 9.4) AMOC (4.6.3.2; 8.6.1; Yes, Medium Reversible within centuries, Very likely decline; Medium confidence of no More lines of evidence 9.2.3.1) confidence High confidence collapse than SROCC Southern MOC (9.2.3.2) Yes, Medium Reversible within decades to Medium confidence in decrease in strength More lines of evidence confidence centuries, Low confidence than SROCC Ocean Acidification Yes, High Reversible at surface; Virtually certain to continue with increasing CO2; More lines of evidence (4.3.2.5; 5.4.2 ; 5.4.4) confidence irreversible for centuries to Likely polar aragonite undersaturation than SROCC millennia at depth, Very high confidence Ocean Deoxygenation Yes, High Reversible at surface; Medium confidence in deoxygenation rates and Improved CMIP6 (5.3.3.2) confidence irreversible for centuries to increased hypoxia simulation millennia at depth, Medium confidence 1 2 [END TABLE 4.10 HERE] 3 4 5 4.8 Low-Likelihood High-Warming Storylines 6 7 Previous IPCC assessments have primarily assessed the projected likely range of changes (e.g., (Collins et 8 al., 2013), see also BOX 1.1). The focus on the likely range partly results from the design of model 9 intercomparison projects that are not targeted to systematically assess the upper and lower bounds of 10 projections, which in principle would require a systematic sampling of structural and parametric model 11 uncertainties. The upper and lower bounds of model projections may further be sensitive to the missing 12 representation of processes and to deep uncertainties about aspects of the climate system (Section 1.2.3.1). 13 14 However, a comprehensive risk assessment requires taking into account also high potential levels of 15 warming whose likelihood is low, but potential impacts on society and ecosystems are high (Xu and 16 Ramanathan, 2017a; Sutton, 2018). Climate-related risks have been argued to increase with increasing levels 17 of global warming even if their likelihood decreases (O’Neill et al., 2017). Thus, it has recently been argued 18 that an assessment that is too narrowly focused on the likely range potentially ignores the changes in the 19 physical climate system associated with the highest risks ((Sutton, 2018), see Section 1.4.4.1). 20 Do Not Cite, Quote or Distribute 4-97 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Given that the CMIP experiments can be considered ensembles of opportunity that are not designed for 2 probabilistic assessments, alternative approaches such as physically plausible high-impact scenarios (Sutton, 3 2018) or storylines have been suggested to investigate the tail of the distribution (Lenderink et al., 2014; 4 Zappa and Shepherd, 2017; Kjellström et al., 2018; Shepherd et al., 2018) (see Section 1.4.4). Such 5 storylines informed by a combination of process understanding, model evidence, and paleo information can 6 be used for risk assessment and adaptation planning to test how well adaptation strategies would cope if the 7 impacts of climate change were more severe than suggested by the likely model range (see Chapter 1 Section 8 1.4.4). Note that by definition the lower bound of the likely model range (see Box 4.1) is equally likely as the 9 upper bound. However, low-warming storylines are not specifically assessed in this section to focus on 10 storylines associated with highest risks. This section further focuses on storylines of high and very high 11 global warming levels along with their manifestation in global patterns of temperature and precipitation 12 changes. However, this does not account for the largest potential changes at regional levels, which would 13 require taking into account storylines of regional changes dependent on changes in atmospheric circulation, 14 land-atmosphere interactions, and regional to local feedbacks. 15 16 This section adopts an approach suggested in Sutton (2018). Since changes in temperature and precipitation 17 tend to increase with the level of warming (Section 4.6.1), low-likelihood high-warming storylines are here 18 illustrated for a level of warming consistent with the upper bound of the assessed very likely range (see 19 Section 4.3.4) and for a level of warming above the very likely range. ECS and TCR are the dominant 20 sources of uncertainty in projections of future warming under moderate to strong emission scenarios (Section 21 7.5.7). Thus, a very high level of warming may occur if ECS and TCR are close to or above the upper bound 22 of the assessed very likely range, which, to agree with historical trends, would require a strong historical 23 aerosol cooling and/or strong SST pattern effects, combined with strong positive cloud feedback and 24 substantial biases in paleoclimate temperature reconstructions, each of which are assessed as either unlikely 25 or very unlikely, though not ruled out (Section 7.5.5). 26 27 For SSP1-2.6, the warming consistent with the upper bound of the assessed very likely range corresponds to 28 a warming of 1.5°C in 2081–2100 relative to 1995–2014 and 2.4°C relative to 1850–1900 (Section 4.3.4), a 29 warming well above the 2°C warming level even in SSP1-2.6. Based on different lines of evidence, Figure 30 4.41 illustrates by how much such a low-likelihood high-warming storyline exceeds the warming pattern 31 consistent with the assessed best estimate GSAT warming of 0.9°C relative to 1995–2014. The first estimate 32 (Figure 4.41, second row) is based on the assumption that the multi-model mean temperature pattern scales 33 linearly with global mean warming. While linear scaling provides an appropriate approximation for changes 34 in temperatures patterns at lower levels of warming (Section 4.2.4), this assumption cannot easily be tested 35 for an extrapolation to higher levels of warming. Thus, a second estimate (Figure 4.41, third row) is based on 36 the average of the five models that simulate a GSAT warming most consistent with the upper bound of the 37 assessed very likely range (see Box 4.1 and Section 4.3.4; note some of the models share components). The 38 two estimates for the annual mean temperature pattern for a low-likelihood high-warming storyline 39 consistently show a warming pattern that substantially exceeds the best estimate warming pattern in most 40 regions except around the North Atlantic and the parts of the Arctic. Pattern scaling suggests more than 50% 41 warming above the best estimate, with 2−3°C warming over much of Eurasia and North America and more 42 than 4°C warming relative to 1995–2014 over the Arctic (Figure 4.41c). The other approach based on five 43 models shows less warming than the best estimate and even larger area of cooling in the North Atlantic but 44 more warming than the best estimate over much of the tropical Pacific, Atlantic, around Antarctica and other 45 the land regions (Figure 4.41e). 46 47 For the high-emission scenarios SSP3-7.0 and SSP5-8.5, a high-warming storyline is associated with wide- 48 spread warming that exceeds the already high best-estimate warming by another 35−50%. For SSP5-8.5, this 49 corresponds to a warming of 1°C−3°C in addition to the best estimate over most land regions, which implies 50 more than 6°C relative to 1995–2014 over most extra-tropical land regions and Amazonia. Over large parts 51 of the Arctic, annual mean temperatures increase by more than 10°C relative to 1995–2014 in such a high- 52 warming storyline under SSP5-8.5. The two lines of evidence yield more consistent patterns for SSP5-8.5 53 than for SSP1-2.6, but there are substantial differences concerning whether the strongest warming above the 54 best estimate occurs over the tropics or extratropical land regions. 55 Do Not Cite, Quote or Distribute 4-98 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 While individual models project even stronger warming over extratropical land regions (Figure 4.41 bottom 2 row), their projected GSAT warming exceeds the assessed very likely 5–95% range and thus correspond to 3 an extremely unlikely (below 5% likelihood) storyline. While all the models consistent with such a storyline 4 tend to overestimate the observed warming trend over the historical period (Brunner et al., 2020; Liang et al., 5 2020; Nijsse et al., 2020; Tokarska et al., 2020; Ribes et al., 2021), some of them show a good representation 6 of several aspects of the present-day climate (Andrews et al., 2019; Sellar et al., 2019; Swart et al., 2019). 7 Such a very high-warming storyline implies widespread warming of more than 1.5°C and 3°C above the 8 best-estimate warming pattern under SSP1-2.6 and SSP5-8.5, respectively. Under SSP1-2.6, this corresponds 9 to more than 3°C warming relative to 1995–2014 over land regions in the northern mid- to high latitudes and 10 more than 6°C in the Arctic (Figure 4.41g). Under SSP5-8.5, such a very high-warming storyline implies 11 more than 8°C warming over parts of Amazonia and more than 6°C over most other tropical land regions 12 (Figure 4.41h). 13 14 15 [START FIGURE 4.41 HERE] 16 17 Figure 4.41: High-warming storylines for changes in annual mean temperature. (a, b) Changes in 2081–2100 18 relative to 1995–2014 consistent with the assessed best GSAT estimate (0.9°C and 3.5°C relative to 19 1995–2014 for SSP1-2.6 and SSP5-8.5, respectively). The CMIP6 multi-model mean is linearly pattern- 20 scaled to the best GSAT estimate. (c–h) Annual mean warming above the best estimate (relative to panels 21 a and b, respectively, note the different colour bar) in a high and very high-warming storyline for 2081– 22 2100. (c, d) Multi-model mean warming pattern scaled to very high GSAT level corresponding to the 23 upper bound of the assessed very likely range (4.8°C for SSP5-8.5 and 1.5°C for SSP1-2.6, see Section 24 4.3.4). (e, f) Average of five models with high GSAT warming nearest to the upper estimate of the very 25 likely range (CESM2, CESM2-WACCM, CNRM-CM6-1, CNRM-CM6-1-HR, EC-Earth3 for SSP1-2.6 26 and ACCESS-CM2, CESM2, CESM2-WACCM, CNRM-CM6-1, CNRM-CM6-1-HRfor SSP5-8.5), (g, 27 h) Average of four and five models, respectively (ACCESS-CM2, HadGEM3-GC31-LL, HadGEM3- 28 GC31-MM, UKESM1-0-LL for SSP1-2.6 and CanESM5, CanESM5-CanOE, HadGEM3-GC31-LL: 29 HadGEM3-GC31-MM, UKESM1-0-LL for SSP5-8.5) projecting very high GSAT warming exceeding 30 the very likely range. Further details on data sources and processing are available in the chapter data table 31 (Table 4.SM.1). 32 33 [END FIGURE 4.41 HERE] 34 35 36 High-warming storylines are very likely also associated with substantial changes in the hydrological cycle 37 due to strong thermodynamic changes, which can be amplified or offset by dynamical changes (Emori and 38 Brown, 2005; Seager et al., 2014b; Chavaillaz et al., 2016b; Kröner et al., 2017; Chen et al., 2019). Here the 39 assessment of the hydrological cycle in high-warming storylines is limited to changes in annual mean 40 precipitation, but changes in seasonal mean precipitation can be even stronger due to enhanced seasonality in 41 many regions (Chapter 8, Box 8.2). 42 43 Quantifying precipitation changes associated with high-warming storylines is challenging since models show 44 the largest changes in precipitation over different regions (Sections 4.5.1 and 4.6.1). In some areas, models 45 project opposing signals in different seasons or a combination of decreasing mean and increasing extreme 46 precipitation (Kendon et al., 2014; Ban et al., 2015; Giorgi et al., 2016; Pendergrass et al., 2017). Models 47 with the most pronounced GSAT warming are not necessarily associated with the strongest precipitation 48 response in all regions, in part due to projected changes in atmospheric dynamics (Madsen et al., 2017; 49 Zappa and Shepherd, 2017; Li et al., 2018). 50 51 Different alternative estimates of changes in annual mean precipitation patterns consistent with high- 52 warming levels are compared here. The first estimate (Figure 4.42b) is based on a linear pattern scaling of 53 the multi-model mean precipitation pattern for SSP5-8.5 (Figure 4.42a) to be consistent with the upper 54 bound of the assessed very likely GSAT range (see above). This estimate is reasonably consistent with the 55 average response of the five models with GSAT warming most consistent with the upper bound of the very 56 likely warming range (Figure 4.42c) except for Australia. Both estimates show about 30−40% larger changes Do Not Cite, Quote or Distribute 4-99 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 in annual mean precipitation than the response pattern consistent with the best GSAT estimate. In a high- 2 warming storyline, widespread increases of more than 30% occur in many regions north of 50°N and over 3 parts of the tropics. Around the Mediterranean and other parts of the subtropics, a high-warming storyline is 4 associated with a reduction in annual mean precipitation of more than 30% depending on the season. 5 6 Both the multi-model mean and the pattern-scaled responses show a smoother pattern than in individual 7 simulations (Tebaldi and Knutti, 2007; Knutti et al., 2010), because the multi-model mean filters out internal 8 variability and because model differences in the location of the largest change tend to cancel. Individual 9 model simulations show opposing signs in precipitation change such as over parts of Australia, the west 10 coast of North America, parts of West Africa and India (Figure 4.42d), which tend to offset in the multi- 11 model mean response. The spatial probability distribution of precipitation changes shows that areas of strong 12 precipitation increase or decrease occur in all models (Figure 4.42g, see also Section 4.6.1). However, due to 13 the spatial smoothing, the multi-model mean response shows a lower area fraction of drying than most of the 14 individual models (Tebaldi and Knutti, 2007; Knutti et al., 2010). The five models with GSAT warming 15 consistent with a high-warming storyline and the two models projecting GSAT warming exceeding the very 16 likely GSAT warming range show a much larger area fraction of drying and somewhat larger fraction of 17 strong precipitation increases than the multi-model mean (Figure 4.42 b–d). 18 19 The high-warming storyline shown in Figure 4.42b, c does not correspond to an upper or lower estimate of 20 annual precipitation increase and decrease over individual locations, which in many regions may differ in the 21 sign of the response (Figure 4.42e, f) due to differences in the model response and internal variability 22 (Madsen et al., 2017). Figure 4.42e, f illustrates upper and lower local estimates corresponding to the 5–95% 23 model range of local uncertainties as opposed to the global-warming storylines. Note, however, that Figure 24 4.42e, f does not show a physically plausible global precipitation response pattern, because information at 25 the different grid points is taken from different model simulations. 26 27 Again, the manifestation of changes in the hydrological cycle for a high-warming storyline is not limited to 28 precipitation, but would substantially affect other variables such as soil moisture, runoff, atmospheric 29 humidity, and evapotranspiration. The changes are also not limited to annual mean precipitation but may be 30 stronger or weaker for individual seasons and for precipitation extremes and dry spells. 31 32 While this assessment is limited to temperature and precipitation, such a high-warming storyline would 33 manifest itself also in other climate variables (Sanderson et al., 2011) assessed in this chapter such as Arctic 34 sea ice, atmospheric circulation changes, and sea-level rise (Ramanathan and Feng, 2008; Xu and 35 Ramanathan, 2017b; Steffen et al., 2018). 36 37 In summary, while high-warming storylines – those associated with global warming levels above the upper 38 bound of the assessed very likely range – are by definition extremely unlikely, they cannot be ruled out. For 39 SSP1-2.6, such a high-warming storyline implies warming well above rather than well below 2°C (high 40 confidence). Irrespective of scenario, high-warming storylines imply changes in many aspects of the climate 41 system that exceed the patterns associated with the best estimate of GSAT changes by up to more than 50% 42 (high confidence). 43 44 45 [START FIGURE 4.42 HERE] 46 47 Figure 4.42: High-warming storylines for changes in annual mean precipitation. (a) Estimates for annual mean 48 precipitation changes in 2081–2100 relative 1995–2014, consistent with the best GSAT estimate derived 49 by linearly scaling the CMIP6 multi-model mean changes to a GSAT change of 3.5°C. (b, c) Estimates 50 for annual mean precipitation changes in 2081–2100 relative 1995–2014 in a storyline representing a 51 physically plausible high-global-warming level. (b) Multi-model mean precipitation scaled to high- 52 global-warming level (corresponding to 4.8°C, the upper bound of the very likely range, see Section 53 4.3.4). (c) Average of five models with GSAT warming nearest to the high level of warming (ACCESS- 54 CM2, CESM2, CESM2-WACCM, CNRM-CM6-1, CNRM-CM6-1-HR) (d) Annual mean precipitation 55 changes in four of the five individual model simulations averaged in (c). (e, f) Local upper estimate (95% 56 quantile across models) and lower estimate (5% quantile across models) at each grid point. Information at Do Not Cite, Quote or Distribute 4-100 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 individual grid points comes from different model simulations and illustrates local uncertainty range but 2 should not be interpreted as a pattern. (g) Area fraction of changes in annual mean precipitation 2081– 3 2100 relative to 1995–2014 for all CMIP6 model simulations (thin black lines), models shown in (c) (red 4 lines), and models showing very high warming above the models shown in (c). The grey range illustrates 5 the 5–95% range across CMIP6 models and the solid black line the area fraction of the multi-model mean 6 pattern shown in (a). Further details on data sources and processing are available in the chapter data table 7 (Table 4.SM.1). 8 9 [END FIGURE 4.42 HERE] 10 Do Not Cite, Quote or Distribute 4-101 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Frequently Asked Questions 2 3 FAQ 4.1: How Will the Climate Change over the Next Twenty Years? 4 5 The parts of the climate system that have shown clear increasing or decreasing trends in recent decades will 6 continue these trends for at least the next twenty years. Examples include changes in global surface 7 temperature, Arctic sea ice cover, and global average sea level. However, over a period as short as twenty 8 years, these trends are substantially influenced by natural climate variability, which can either amplify or 9 attenuate the trend expected from the further increase in greenhouse gas concentrations. 10 11 Twenty years are a long time by human standards but a short time from a climate point of view. Emissions of 12 greenhouse gases will continue over the next twenty years, as assumed in all the scenarios considered in this 13 report, albeit with varying rates. These emissions will further increase concentrations of greenhouse gases in 14 the atmosphere (see FAQ 4.2), leading to continued trends in global surface warming and other parts of the 15 climate system, including Arctic sea ice and global average sea level (see FAQ 9.2). FAQ 4.1, Figure 1 16 shows that both global surface temperature rise and the shrinking of sea ice in the Arctic will continue, with 17 little difference between high- and low-emission scenarios over the next 20 years (that is, between the red 18 and blue lines). 19 20 However, these expected trends will be overlain by natural climate variability (see FAQ 3.2). First, a major 21 volcanic eruption might occur, such as the 1991 eruption of Mt. Pinatubo on the Philippines; such an 22 eruption might cause a global surface cooling of a few tenths of a degree Celsius lasting several years. 23 Second, both atmosphere and ocean show variations that occur spontaneously, without any external 24 influence. These variations range from localized weather systems to continent- and ocean-wide patterns and 25 oscillations that change over months, years, or decades. Over a period of twenty years, natural climate 26 variability strongly influences many climate quantities, when compared to the response to the increase in 27 greenhouse gas concentrations from human activities. The effect of natural variability is illustrated by the 28 very different trajectories that individual black, red or blue lines can take in FAQ 4.1, Figure 1. Whether 29 natural variability would amplify or attenuate the human influence cannot generally be predicted out to 30 twenty years into the future. Natural climate variability over the next twenty years thus constitutes an 31 uncertainty that at best can be quantified accurately but that cannot be reduced. 32 33 Locally, the effect of natural variability would be much larger still. Simulations (not shown here) indicate 34 that, locally, a cooling trend over the next twenty year cannot be ruled out, even under the high-emission 35 scenario – at a small number of locations on Earth, but these might lie anywhere. Globally, though, 36 temperatures would rise under all scenarios. 37 38 In summary, while the direction of future change is clear for the two important climate quantities shown here 39 − the global surface temperature and the Arctic sea-ice area in September − the magnitude of the change is 40 much less clear because of natural variability. 41 42 43 [START FAQ 4.1, FIGURE 1 HERE] 44 45 FAQ 4.1, Figure 1: Simulations over the period 1995–2040, encompassing the recent past and the next twenty 46 years, of two important indicators of global climate change, (top) global surface temperature, 47 and (bottom), the area of Arctic sea ice in September. Both quantities are shown as deviations 48 from the average over the period 1995–2014. The black curves are for the historical period ending 49 in 2014; the blue curves represent a low-emission scenario (SSP1-2.6) and the red curves one 50 high-emission scenario (SSP3-7.0). 51 52 53 [END FAQ 4.1, FIGURE 1 HERE] 54 55 56 FAQ 4.2: How Quickly Would We See the Effects of Reducing Carbon Dioxide Emissions? Do Not Cite, Quote or Distribute 4-102 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 2 The effects of substantial reductions in carbon dioxide emissions would not be apparent immediately, and the 3 time required to detect the effects would depend on the scale and pace of emissions reductions. Under the 4 lower-emission scenarios considered in this report, the increase in atmospheric carbon dioxide 5 concentrations would slow visibly after about five to ten years, while the slowing down of global surface 6 warming would be detectable after about twenty to thirty years. The effects on regional precipitation trends 7 would only become apparent after several decades. 8 9 Reducing emissions of carbon dioxide (CO2) – the most important greenhouse gas emitted by human 10 activities – would slow down the rate of increase in atmospheric CO2 concentration. However, 11 concentrations would only begin to decrease when net emissions approach zero, that is, when most or all of 12 the CO2 emitted into the atmosphere each year is removed by natural and human processes (see FAQ 5.1, 13 FAQ 5.3). This delay between a peak in emissions and a decrease in concentration is a manifestation of the 14 very long lifetime of CO2 in the atmosphere; part of the CO2 emitted by humans remains in the atmosphere 15 for centuries to millennia. 16 17 Reducing the rate of increase in CO2 concentration would slow down global surface warming within a 18 decade. But this reduction in the rate of warming would initially be masked by natural climate variability and 19 might not be detected for a few decades (see FAQ 1.2, FAQ 3.2, FAQ 4.1). Detecting whether surface 20 warming has indeed slowed down would thus be difficult in the years right after emissions reductions begin. 21 22 The time needed to detect the effect of emissions reductions is illustrated by comparing low- and high- 23 emission scenarios (FAQ 4.2, Figure 1). In the low-emission scenario (SSP1-2.6), CO2 emissions level off 24 after 2015 and begin to fall in 2020, while they keep increasing throughout the 21st century in the high- 25 emission scenario (SSP3-7.0). The uncertainty arising from natural internal variability in the climate system 26 is represented by simulating each scenario ten times with the same climate model but starting from slightly 27 different initial states back in 1850 (thin lines). For each scenario, the differences between individual 28 simulations are caused entirely by simulated natural internal variability. The average of all simulations 29 represents the climate response expected for a given scenario. The climate history that would actually unfold 30 under each scenario would consist of this expected response combined with the contribution from natural 31 internal variability and the contribution from potential future volcanic eruptions (the latter effect is not 32 represented here). 33 34 FAQ 4.2, Figure 1 shows that the atmospheric CO2 concentrations differ noticeably between the two 35 scenarios about five to ten years after the emissions have begun to diverge in year 2015. In contrast, the 36 difference in global surface temperatures between the two scenarios does not become apparent until later – 37 about two to three decades after the emissions histories have begun to diverge in this example. This time 38 would be longer if emissions were reduced more slowly than in the low-emission scenario illustrated here 39 and shorter in the case of stronger reductions. Detection would take longer for regional quantities and for 40 precipitation changes, which vary more strongly from natural causes. For instance, even in the low-emission 41 scenario, the effect of reduced CO2 emissions would not become visible in regional precipitation until late in 42 the 21st century. 43 44 In summary, it is only after a few decades of reducing CO2 emissions that we would clearly see global 45 temperatures starting to stabilise. By contrast, short-term reductions in CO2 emissions, such as during the 46 COVID-19 pandemic, do not have detectable effects on either CO2 concentration or global temperature. 47 Only sustained emission reductions over decades would have a widespread effect across the climate system. 48 49 50 [START FAQ 4.2, FIGURE 1 HERE] 51 52 FAQ 4.2, Figure 1: Observing the benefits of emission reductions. (top) Carbon dioxide (CO2) emissions, (middle) 53 CO2 concentration in the atmosphere and (bottom) effect on global surface temperature for two scenarios: a 54 low-emission scenario (SSP1-2.6, blue) and a high-emission scenario (SSP3-7.0). In the low-emission 55 scenario, CO2 emissions begin to decrease in 2020 whereas they keep increasing throughout the 21st Do Not Cite, Quote or Distribute 4-103 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 century in the high-emission scenario. The thick lines are the average of the ten individual simulations (thin 2 line) for each scenario. Differences between individual simulations reflect natural variability. 3 4 [END FAQ 4.2, FIGURE 1 HERE] 5 6 7 FAQ 4.3: At a given level of global warming, what are the spatial patterns of climate change? 8 9 As the planet warms, climate change does not unfold uniformly across the globe, but some patterns of 10 regional change show clear, direct and consistent relationships to increases in global surface temperature. 11 The Arctic warms more than other regions, land areas warm more than the ocean surface, and the Northern 12 Hemisphere more than the Southern Hemisphere. Precipitation increases over high latitudes, tropics and 13 large parts of the monsoon regions, but decreases over the subtropics. For cases like these, we can infer the 14 direction and magnitude of some regional changes – particularly temperature and precipitation changes – 15 for any given level of global warming. 16 17 The intensity of climate change will depend on the level of global warming. It is possible to identify certain 18 patterns of regional climate change that occur consistently, but increase in amplitude, across increasing 19 levels of global warming. Such robust spatial patterns of climate change are largely independent of the 20 specific scenario (and pathway in time) that results in a given level of global warming. That is, as long as 21 different scenarios result in the same global warming level, irrespective of the time when this level is 22 attained in each scenario, we can infer the patterns of regional change that would result from this warming. 23 When patterns of changes are robust, regional consequences can be assessed for all levels of global warming, 24 for all future time periods, and for all scenarios. Temperature and precipitation show such robust patterns of 25 changes that are particularly striking. 26 27 The high latitudes of the Northern Hemisphere are projected to warm the most, by two to four times the level 28 of global warming – a phenomenon referred to as Arctic amplification (FAQ 4.3 Figure 1, left). Several 29 processes contribute to this high rate of warming, including increases in the absorption of solar radiation due 30 to the loss of reflective sea ice and snow in a warmer world. In the Southern Hemisphere, Antarctica is 31 projected to warm faster than the mid-latitude Southern Ocean, but the Southern Hemisphere high latitudes 32 are projected to warm at a reduced amplitude compared to the level of global warming (FAQ 4.3 Figure 1, 33 left). An important reason for the relatively slower warming of the Southern Hemisphere high latitudes is the 34 upwelling of Antarctic deep waters that drives a large surface heat uptake in the Southern Ocean. 35 36 The warming is generally stronger over land than over the ocean, and in the Northern Hemisphere compared 37 to the Southern Hemisphere, and with less warming over the central subpolar North Atlantic and the 38 southernmost Pacific. The differences are the result of several factors, including differences in how land and 39 ocean areas absorb and retain heat, the fact that there is more land area in the Northern Hemisphere than in 40 the Southern Hemisphere, and the influence of ocean circulation. In the Southern Hemisphere, robust 41 patterns of relatively high warming are projected for subtropical South America, southern Africa, and 42 Australia. The relatively strong warming in subtropical southern Africa arises from strong interactions 43 between soil moisture and temperature and from increased solar radiation as a consequence of enhanced 44 subsidence. 45 46 Precipitation changes are also proportional to the level of global warming (FAQ 4.3 Figure 1, right), 47 although uncertainties are larger than for the temperature change. In the high latitudes of both the Southern 48 and Northern Hemispheres, increases in precipitation are expected as the planet continues to warm, with 49 larger changes expected at higher levels of global warming (FAQ 4.3 Figure 1, right). The same holds true 50 for the projected precipitation increases over the tropics and large parts of the monsoon regions. General 51 drying is expected over the subtropical regions, particularly over the Mediterranean, southern Africa and 52 parts of Australia, South America, and southwest North America, as well as over the subtropical Atlantic and 53 parts of the subtropical Indian and Pacific Oceans. Increases in precipitation over the tropics and decreases 54 over the subtropics amplify with higher levels of global warming. 55 Do Not Cite, Quote or Distribute 4-104 Total pages: 195 Final Government Distribution Chapter 4 IPCC AR6 WGI 1 Some regions that are already dry and warm, such as southern Africa and the Mediterranean, are expected to 2 become progressively drier and drastically warmer at higher levels of global warming. 3 4 In summary, climate change will not affect all the parts of the globe evenly. Rather, distinct regional patterns 5 of temperature and precipitation change can be identified, and these changes are projected to amplify as the 6 level of global warming increases. 7 8 9 [START FAQ 4.3, FIGURE 1 HERE] 10 11 FAQ 4.3, Figure 1: Regional changes in temperature (left) and precipitation (right) are proportional to the level 12 of global warming, irrespective of the scenario through which the level of global warming is 13 reached. Surface warming and precipitation change are shown relative to the 1850–1900 climate, 14 and for time periods over which the globally averaged surface warming is 1.5°C (top) and 3°C 15 (bottom), respectively. Changes presented here are based on thirty-one CMIP6 models using the 16 high-emission scenario SSP3-7.0. 17 18 [END FAQ 4.3, FIGURE 1 HERE] 19 20 21 Do Not Cite, Quote or Distribute 4-105 Total pages: 195