Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Table of Contents 2 Executive Summary .......................................................................................................................................... 5 3 9.1 Introduction .................................................................................................................................. 11 4 BOX 9.1: Key processes driving sea level change .................................................................................... 12 5 9.2 Oceans ........................................................................................................................................... 14 6 9.2.1 Ocean surface ............................................................................................................................... 14 7 9.2.1.1 Sea Surface Temperature (SST) ............................................................................................. 14 8 9.2.1.2 Air-sea fluxes ............................................................................................................................ 16 9 9.2.1.3 Upper Ocean Stratification and Surface Mixed Layers ....................................................... 18 10 BOX 9.2: Marine Heatwaves .................................................................................................................... 20 11 9.2.2 Changes in Heat and Salinity ...................................................................................................... 21 12 9.2.2.1 Ocean Heat Content and Heat Transport ............................................................................. 21 13 9.2.2.2 Ocean Salinity .......................................................................................................................... 26 14 9.2.2.3 Water Masses ........................................................................................................................... 28 15 9.2.3 Regional Ocean Circulation ........................................................................................................ 30 16 9.2.3.1 Atlantic Meridional Overturning Circulation....................................................................... 30 17 9.2.3.2 Southern Ocean ........................................................................................................................ 33 18 9.2.3.3 Tropical Oceans ....................................................................................................................... 36 19 9.2.3.4 Gyres, Western Boundary Currents, and Inter-Basin Exchanges ...................................... 36 20 9.2.3.5 Eastern Boundary Upwelling Systems ................................................................................... 39 21 9.2.3.6 Coastal Systems and Marginal Seas ....................................................................................... 40 22 9.2.4 Steric and dynamic sea-level change .......................................................................................... 41 23 9.2.4.1 Global mean thermosteric sea-level change .......................................................................... 41 24 9.2.4.2 Ocean dynamic sea-level change ............................................................................................ 42 25 9.3 Sea ice ............................................................................................................................................ 44 26 9.3.1 Arctic Sea Ice ................................................................................................................................ 44 27 9.3.1.1 Arctic Sea-Ice Coverage .......................................................................................................... 44 28 9.3.1.2 Arctic Sea-Ice volume and thickness ...................................................................................... 48 29 9.3.2 Antarctic Sea Ice .......................................................................................................................... 49 30 9.3.2.1 Antarctic sea-ice coverage ....................................................................................................... 49 31 9.3.2.2 Antarctic sea-ice thickness ...................................................................................................... 51 32 9.4 Ice Sheets....................................................................................................................................... 52 Do Not Cite, Quote or Distribute 9-2 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 9.4.1 Greenland Ice Sheet ..................................................................................................................... 52 2 9.4.1.1 Recent observed changes ......................................................................................................... 52 3 9.4.1.2 Model evaluation ...................................................................................................................... 56 4 9.4.1.3 Projections to 2100 ................................................................................................................... 57 5 9.4.1.4 Projections beyond 2100.......................................................................................................... 61 6 BOX 9.3: Insights into land ice evolution from model intercomparison projects................................... 62 7 9.4.2 Antarctic Ice Sheet ....................................................................................................................... 64 8 9.4.2.1 Recent observed changes ......................................................................................................... 64 9 9.4.2.2 Model evaluation ...................................................................................................................... 67 10 9.4.2.3 Drivers of future Antarctic ice sheet change ......................................................................... 69 11 9.4.2.4 Ice sheet instabilities ................................................................................................................ 71 12 9.4.2.5 Projections to 2100 ................................................................................................................... 72 13 9.4.2.6 Projections beyond 2100.......................................................................................................... 76 14 9.5 Glaciers, permafrost and seasonal snow cover .......................................................................... 78 15 9.5.1 Glaciers ......................................................................................................................................... 78 16 9.5.1.1 Observed and reconstructed glacier extent and mass changes ............................................ 78 17 9.5.1.2 Model evaluation ...................................................................................................................... 82 18 9.5.1.3 Projections ................................................................................................................................ 83 19 9.5.2 Permafrost .................................................................................................................................... 85 20 9.5.2.1 Observed and reconstructed changes..................................................................................... 85 21 9.5.2.2 Evaluation of permafrost in climate models .......................................................................... 87 22 9.5.2.3 Projected permafrost changes ................................................................................................ 89 23 9.5.3 Seasonal snow cover ..................................................................................................................... 89 24 9.5.3.1 Observed changes of seasonal snow cover ............................................................................. 90 25 9.5.3.2 Evaluation of seasonal snow in climate models ..................................................................... 92 26 9.5.3.3 Projected snow cover changes ................................................................................................ 94 27 9.6 Sea Level Change ......................................................................................................................... 94 28 9.6.1 Global and regional sea-level change in the instrumental era ................................................. 94 29 9.6.1.1 Global mean sea-level change budget in the pre-satellite era .............................................. 94 30 9.6.1.2 Global mean sea-level change budget in the satellite era ..................................................... 96 31 9.6.1.3 Regional sea-level change in the satellite era......................................................................... 98 32 9.6.1.4 Attribution and time of emergence of regional sea-level change ......................................... 99 Do Not Cite, Quote or Distribute 9-3 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Cross-Chapter Box 9.1: Global energy inventory and sea level budget .................................................... 100 2 9.6.2 Paleo context of global and regional sea-level change ............................................................ 101 3 9.6.3 Future sea-level changes ............................................................................................................ 106 4 9.6.3.1 Global mean sea level projections based on the Representative Concentration Pathways ... 5 ................................................................................................................................................. 106 6 9.6.3.2 Drivers of projected sea-level change................................................................................... 108 7 9.6.3.3 Sea-level projections to 2150 based on SSP scenarios ........................................................ 115 8 9.6.3.4 Sea-level projections up to 2100 based on global warming levels...................................... 118 9 9.6.3.5 Multi-century and multi-millennial sea-level rise ............................................................... 119 10 BOX 9.4: High-end storyline of 21st century sea-level rise ................................................................... 122 11 9.6.4 Extreme sea levels: Tides, surges and waves ........................................................................... 124 12 9.6.4.1 Past changes ........................................................................................................................... 124 13 9.6.4.2 Future changes ....................................................................................................................... 126 14 9.7 Final Remarks ............................................................................................................................ 129 15 Frequently Asked Questions ........................................................................................................................ 131 16 Acknowledgements ....................................................................................................................................... 137 17 Reference ...................................................................................................................................................... 138 18 Figures .......................................................................................................................................................... 202 19 20 21 22 Do Not Cite, Quote or Distribute 9-4 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Executive Summary 2 3 This chapter assesses past and projected changes in the ocean, cryosphere and sea level using paleo- 4 reconstructions, instrumental observations and model simulations. In the following summary, we update and 5 expand the related assessments from the IPCC Fifth Assessment Report (AR5), the Special Report on Global 6 Warming of 1.5ºC (SR1.5) and the Special Report on Ocean and Cryosphere in a Changing Climate 7 (SROCC). Major advances in this chapter since the SROCC include the synthesis of extended and new 8 observations, which allows for improved assessment of past change, processes and budgets for the last 9 century, and the use of a hierarchy of models and emulators, which provide improved projections and 10 uncertainty estimates of future change. In addition, the systematic use of model emulators makes our 11 projections of ocean heat content, land-ice loss and sea level rise fully consistent both with each other and 12 with the assessed equilibrium climate sensitivity and projections of global surface air temperature across the 13 entire report. In this executive summary, uncertainty ranges are reported as very likely ranges and expressed 14 by square brackets, unless otherwise noted. 15 16 Ocean Heat and Salinity 17 18 At the ocean surface, temperature has on average increased by 0.88 [0.68–1.01] °C from 1850-1900 to 19 2011-2020, with 0.60 [0.44–0.74] °C of this warming having occurred since 1980. The ocean surface 20 temperature is projected to increase from 1995–2014 to 2081–2100 on average by 0.86 [0.43–1.47, 21 likely range] °C in SSP1-2.6 and by 2.89 [2.01–4.07, likely range] °C in SSP5-8.5. Since the 1950s, the 22 fastest surface warming has occurred in the Indian Ocean and in Western Boundary Currents, while ocean 23 circulation has caused slow warming or surface cooling in the Southern Ocean, equatorial Pacific, North 24 Atlantic, and coastal upwelling systems (very high confidence). At least 83% of the ocean surface will very 25 likely warm over the 21st century in all SSP scenarios. {2.3.3, 9.2.1} 26 27 The heat content of the global ocean has increased since at least 1970 and will continue to increase 28 over the 21st century (virtually certain). The associated warming will likely continue until at least 2300 29 even for low-emission scenarios because of the slow circulation of the deep ocean. Ocean heat content 30 has increased from 1971 to 2018 by [0.28–0.55] yottajoules and will likely increase until 2100 by 2 to 4 31 times that amount under SSP1-2.6 and 4 to 8 times that amount under SSP5-8.5. The long time scale also 32 implies that the amount of deep-ocean warming will only become scenario-dependent after about 2040 and 33 that the warming is irreversible over centuries to millennia (medium confidence). On annual to decadal time 34 scales, the redistribution of heat by the ocean circulation dominates spatial patterns of temperature change 35 (high confidence). At longer time scales, the spatial patterns are dominated by additional heat primarily 36 stored in water-masses formed in the Southern Ocean, and by weaker warming in the North Atlantic where 37 heat redistribution caused by changing circulation counteracts the additional heat input through the surface 38 (high confidence). {9.2.2, 9.2.4, 9.6.1, Cross-Chapter Box 9.1} 39 40 Marine heatwaves – sustained periods of anomalously high near-surface temperatures that can lead to 41 severe and persistent impacts on marine ecosystems – have become more frequent over the 20th 42 century (high confidence). Since the 1980s, they have approximately doubled in frequency (high 43 confidence) and have become more intense and longer (medium confidence). This trend will continue, 44 with marine heatwaves at global scale becoming 4 [2–9, likely range] times more frequent in 2081–2100 45 compared to 1995–2014 under SSP1-2.6, and 8 [3–15, likely range] times more frequent under SSP5-8.5. 46 The largest changes will occur in the tropical ocean and the Arctic (medium confidence). {Box 9.2} 47 48 The upper ocean has become more stably stratified since at least 1970 over the vast majority of the 49 globe (virtually certain), primarily due to surface-intensified warming and high-latitude surface 50 freshening (very high confidence). Changes in ocean stability affect vertical exchanges of surface waters 51 with the deep ocean and large-scale ocean circulation. Based on recent refined analyses of the available Do Not Cite, Quote or Distribute 9-5 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 observations, the global 0–200 m stratification is now assessed to have increased about twice as much as 2 reported by the SROCC, with a 4.9 ± 1.5% increase from 1970 to 2018 (high confidence) and even higher 3 increases at the base of the surface mixed layer. Upper-ocean stratification will continue to increase 4 throughout the 21st century (virtually certain). {9.2.1} 5 6 Ocean Circulation 7 8 The Atlantic Meridional Overturning Circulation (AMOC) will very likely decline over the 21st 9 century for all SSP scenarios. There is medium confidence that the decline will not involve an abrupt 10 collapse before 2100. For the 20th century, there is low confidence in reconstructed and modelled AMOC 11 changes because of their low agreement in quantitative trends. The low confidence also arises from new 12 observations that indicate missing key processes in both models and measurements used for formulating 13 proxies and from new evaluations of modelled AMOC variability. This results in low confidence in 14 quantitative projections of AMOC decline in the 21st century, despite the high confidence in the future 15 decline as a qualitative feature based on process understanding. {9.2.3} 16 17 Southern Ocean circulation and associated temperature changes in Antarctic ice-shelf cavities are 18 sensitive to changes in wind patterns and increased ice-shelf melt (high confidence). However, 19 limitations in understanding feedback mechanisms involving the ocean, atmosphere and cryosphere, which 20 are not fully represented in the current generation of climate models, generally limit our confidence in future 21 projections of the Southern Ocean and of its forcing on Antarctic sea ice and ice shelves. {9.2.3, 9.3.2, 9.4.2} 22 23 Many ocean currents will change in the 21st century as a response to changes in wind stress associated 24 with anthropogenic warming (high confidence). Western boundary currents have shifted poleward since 25 1993 (medium confidence), consistent with a poleward shift of the subtropical gyres. Of the four eastern 26 boundary upwelling systems, only the California current system has experienced some large-scale 27 upwelling-favourable wind intensification since the 1980s (medium confidence). In the 21st century, 28 consistent with projected changes in the surface winds, the East Australian Current Extension and Agulhas 29 Current Extension will intensify, while the Gulf Stream and Indonesian Throughflow will weaken (medium 30 confidence). Eastern boundary upwelling systems will change, with a dipole spatial pattern within each 31 system of reduction at low latitude and enhancement at high latitude (high confidence). {9.2.1, 9.2.3} 32 33 Sea Ice 34 35 The Arctic Ocean will likely become practically sea ice–free1 during the seasonal sea ice minimum for 36 the first time before 2050 in all considered SSP scenarios. There is no tipping point for this loss of 37 Arctic summer sea ice (high confidence). The practically ice-free state is projected to occur more often 38 with higher greenhouse gas concentrations and will become the new normal for high-emission scenarios by 39 the end of this century (high confidence). Based on observational evidence, Coupled Model Intercomparison 40 Project Phase 6 (CMIP6) models and conceptual understanding, the substantial satellite-observed decrease of 41 Arctic sea ice area over the period 1979–2019 is well described as a linear function of global mean surface 42 temperature, and thus of cumulative anthropogenic CO2 emissions, with superimposed internal variability 43 (high confidence). According to both process understanding and CMIP6 simulations, a practically sea ice– 44 free state will likely be observed in some years before additional (post-2020) cumulative anthropogenic CO2 45 emissions reach 1000 GtCO2. {4.3.2, 9.3.1} 46 1 sea ice area below 1 million km2 Do Not Cite, Quote or Distribute 9-6 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 For Antarctic sea ice, regionally opposing trends and large interannual variability result in no 2 significant trend in satellite-observed sea ice area from 1979 to 2020 in both winter and summer (high 3 confidence). The regionally opposing trends result primarily from changing regional wind forcing (medium 4 confidence). There is low confidence in model simulations of past and future Antarctic sea ice evolution due 5 to deficiencies of process representation, in particular at the regional level. {2.3.2, 9.2.3, 9.3.2} 6 7 Ice Sheets 8 9 The Greenland Ice Sheet has lost 4890 [4140–5640] Gt mass over the period 1992–2020, equivalent to 10 13.5 [11.4–15.6] mm global mean sea level rise. The mass-loss rate was on average 39 [–3 to 80] Gt yr–1 11 over the period 1992–1999, 175 [131 to 220] Gt yr–1 over the period 2000–2009 and 243 [197 to 290] Gt 12 yr–1 over the period 2010–2019. This mass loss is driven by both discharge and surface melt, with the latter 13 increasingly becoming the dominating component of mass loss with high interannual variability in the last 14 decade (high confidence). The largest mass losses occurred in the Northwest and the Southeast of Greenland 15 (high confidence). {2.3.2, 9.4.1} 16 17 The Antarctic Ice Sheet has lost 2670 [1800–3540] Gt mass over the period 1992–2020, equivalent to 18 7.4 [5.0–9.8] mm global mean sea level rise. The mass-loss rate was on average 49 [–2 to 100] Gt yr–1 19 over the period 1992–1999, 70 [22 to 119] Gt yr–1 over the period 2000–2009 and 148 [94 to 202] Gt yr–1 20 over the period 2010–2019. Mass losses from West Antarctic outlet glaciers outpaced mass gain from 21 increased snow accumulation on the continent and dominated the ice sheet mass losses since 1992 (very high 22 confidence). These mass losses from the West Antarctic outlet glaciers were mainly induced by ice shelf 23 basal melt (high confidence) and locally by ice shelf disintegration preceded by strong surface melt (high 24 confidence). Parts of the East Antarctic Ice Sheet have lost mass in the last two decades (high confidence). 25 {2.3.2, 9.4.2, Atlas.11.1} 26 27 Both the Greenland Ice Sheet (virtually certain) and the Antarctic Ice Sheet (likely) will continue to lose 28 mass throughout this century under all considered SSP scenarios. The related contribution to global 29 mean sea level rise until 2100 from the Greenland Ice Sheet will likely be 0.01–0.10 m under SSP 1-2.6, 30 0.04–0.13 m under SSP2-4.5 and 0.09–0.18 m under SSP5-8.5, while the Antarctic Ice Sheet will likely 31 contribute 0.03–0.27 m under SSP1-2.6, 0.03–0.29 m under SSP2-4.5 and 0.03–0.34 m under SSP5-8.5. 32 The loss of ice from Greenland will become increasingly dominated by surface melt, as marine margins 33 retreat and the ocean-forced dynamic response of ice-sheet margins diminishes (high confidence). In the 34 Antarctic, dynamic losses driven by ocean warming and ice shelf disintegration will likely continue to 35 outpace increasing snowfall this century (medium confidence). Beyond 2100, total mass loss from both ice 36 sheets will be greater under high-emission scenarios than under low-emission scenarios (high confidence). 37 The assessed likely ranges consider those ice-sheet processes in whose representation in current models we 38 have at least medium confidence, including surface mass balance and grounding-line retreat in the absence of 39 instabilities. Under high-emission scenarios, poorly understood processes related to Marine Ice Sheet 40 Instability and Marine Ice Cliff Instability, characterized by deep uncertainty, have the potential to strongly 41 increase Antarctic mass loss on century to multi-century time scales. {9.4.1, 9.4.2, 9.6.3, Box 9.3, Box 9.4} 42 43 Glaciers 44 45 Glaciers lost 6200 [4600–7800] Gt of mass (17.1 [12.7–21.5] mm global mean sea level equivalent) over 46 the period 1993 to 2019 and will continue losing mass under all SSP scenarios (very high confidence). 47 During the decade 2010 to 2019, glaciers lost more mass than in any other decade since the beginning 48 of the observational record (very high confidence). For all regions with long-term observations, glacier 49 mass in the decade 2010 to 2019 is the smallest since at least the beginning of the 20th century (medium 50 confidence). Because of their lagged response, glaciers will continue to lose mass at least for several decades 51 even if global temperature is stabilized (very high confidence). Glaciers will lose 29,000 [9,000–49,000] Gt Do Not Cite, Quote or Distribute 9-7 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 and 58,000 [28,000–88,000] Gt over the period 2015–2100 for RCP2.6 and RCP8.5, respectively (medium 2 confidence), which represents 18 [5–31] % and 36 [16–56] % of their early-21st-century mass, respectively. 3 {2.3.2, 9.5.1, 9.6.1, 9.6.3, 12.4} 4 5 Permafrost 6 7 Increases in permafrost temperature have been observed over the past three to four decades 8 throughout the permafrost regions (high confidence), and further global warming will lead to near- 9 surface permafrost volume loss (high confidence). Complete permafrost thaw in recent decades is a 10 common phenomenon in discontinuous and sporadic permafrost regions (medium confidence). Permafrost 11 warmed globally by 0.29 [0.17–0.41, likely range] °C between 2007 and 2016 (medium confidence). An 12 increase in the active layer thickness is a pan-Arctic phenomenon (medium confidence), subject to strong 13 heterogeneity in surface conditions. The volume of perennially frozen soil within the upper 3 m of the 14 ground will decrease by about 25% per 1°C of global surface air temperature change (up to 4°C above pre- 15 industrial temperature) (medium confidence). {9.5.2} 16 17 Snow 18 19 Northern Hemisphere spring snow cover extent has been decreasing since 1978 (very high confidence), 20 and there is high confidence that this trend extends back to 1950. Further decrease of Northern 21 Hemisphere seasonal snow cover extent is virtually certain under further global warming. The observed 22 sensitivity of Northern Hemisphere snow cover extent to Northern Hemisphere land surface air temperature 23 for 1981–2010 is –1.9 [–2.8 to –1.0, likely range] million km2 per 1°C throughout the snow season. It is 24 virtually certain that Northern Hemisphere snow cover extent will continue to decrease as global climate 25 continues to warm, and process understanding strongly suggests that this also applies to Southern 26 Hemisphere seasonal snow cover (high confidence). Northern Hemisphere spring snow cover extent will 27 decrease by about 8% per 1°C of global surface air temperature change (up to 4°C above pre-industrial 28 temperature) (medium confidence). {9.5.3} 29 30 Sea Level 31 32 Global mean sea level (GMSL) rose faster in the 20th century than in any prior century over the last 33 three millennia (high confidence), with a 0.20 [0.15–0.25] m rise over the period 1901 to 2018 (high 34 confidence). GMSL rise has accelerated since the late 1960s, with an average rate of 2.3 [1.6–3.1] mm 35 yr-1 over the period 1971–2018 increasing to 3.7 [3.2–4.2] mm yr-1 over the period 2006–2018 (high 36 confidence). New observation-based estimates published since SROCC lead to an assessed sea level rise 37 over the period 1901 to 2018 that is consistent with the sum of individual components. While ocean thermal 38 expansion (38%) and mass loss from glaciers (41%) dominate the total change from 1901 to 2018, ice sheet 39 mass loss has increased and accounts for about 35% of the sea level increase during the period 2006–2018 40 (high confidence). {2.3.3, 9.6.1, 9.6.2, Cross-Chapter Box 9.1, Box 7.2} 41 42 At the basin scale, sea levels rose fastest in the Western Pacific and slowest in the Eastern Pacific over 43 the period 1993–2018 (medium confidence). Regional differences in sea level arise from ocean dynamics; 44 changes in Earth gravity, rotation and deformation due to land-ice and land-water changes; and vertical land 45 motion. Temporal variability in ocean dynamics dominates regional patterns on annual to decadal time scales 46 (high confidence). The anthropogenic signal in regional sea level change will emerge in most regions by 47 2100 (medium confidence). {9.2.4, 9.6.1} 48 49 Regional sea level change has been the main driver of changes in extreme still water levels across the 50 quasi-global tide gauge network over the 20th century (high confidence) and will be the main driver of Do Not Cite, Quote or Distribute 9-8 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 a substantial increase in the frequency of extreme still water levels over the next century (medium 2 confidence). Observations show that high tide flooding events that occurred five times per year during the 3 period 1960–1980 occurred on average more than eight times per year during the period 1995–2014 (high 4 confidence). Under the assumption that other contributors to extreme sea levels remain constant (e.g., 5 stationary tides, storm-surge, and wave climate), extreme sea levels that occurred once per century in the 6 recent past will occur annually or more frequently at about 19–31% of tide gauges by 2050 and at about 60% 7 (SSP1-2.6) to 82% (SSP5-8.5) of tide gauges by 2100 (medium confidence). In total, such extreme sea levels 8 will occur about 20 to 30 times more frequently by 2050 and 160 to 530 times more frequently by 2100 9 compared to the recent past, as inferred from the median amplification factors for SSP1-2.6, SSP2-4.5, and 10 SSP5-8.5 (medium confidence). Over the 21st century, the majority of coastal locations will experience a 11 median projected regional sea level rise within +/- 20% of the median projected GMSL change (medium 12 confidence). {9.6.4} 13 14 It is virtually certain that global mean sea level will continue to rise through 2100, because all assessed 15 contributors to global mean sea level are likely to virtually certain to continue contributing throughout 16 this century. Considering only processes for which projections can be made with at least medium 17 confidence, relative to the period 1995–2014 GMSL will rise by 2050 between 0.18 [0.15–0.23, likely 18 range] m (SSP1-1.9) and 0.23 [0.20–0.30, likely range] m (SSP5-8.5), and by 2100 between 0.38 [0.28– 19 0.55, likely range] m (SSP1-1.9) and 0.77 [0.63–1.02, likely range] m (SSP5-8.5). This GMSL rise is 20 primarily caused by thermal expansion and mass loss from glaciers and ice sheets, with minor contributions 21 from changes in land-water storage. These likely range projections do not include those ice-sheet-related 22 processes that are characterized by deep uncertainty. {9.6.3} 23 24 Higher amounts of GMSL rise before 2100 could be caused by earlier-than-projected disintegration of 25 marine ice shelves, the abrupt, widespread onset of Marine Ice Sheet Instability and Marine Ice Cliff 26 Instability around Antarctica, and faster-than-projected changes in the surface mass balance and 27 discharge from Greenland. These processes are characterised by deep uncertainty arising from limited 28 process understanding, limited availability of evaluation data, uncertainties in their external forcing and high 29 sensitivity to uncertain boundary conditions and parameters. In a low-likelihood, high-impact storyline, 30 under high emissions such processes could in combination contribute more than one additional meter of sea 31 level rise by 2100. {9.6.3, Box 9.4} 32 33 Beyond 2100, GMSL will continue to rise for centuries due to continuing deep ocean heat uptake and 34 mass loss of the Greenland and Antarctic Ice Sheets, and will remain elevated for thousands of years 35 (high confidence). Considering only processes for which projections can be made with at least medium 36 confidence and assuming no increase in ice-mass flux after 2100, relative to the period 1995–2014, by 2150, 37 GMSL will rise between 0.6 [0.4–0.9, likely range] m (SSP1-1.9) and 1.4 [1.0–1.9, likely range] m (SSP5- 38 8.5). By 2300, GMSL will rise between 0.3 m and 3.1 m under SSP1-2.6, between 1.7 m and 6.8 m under 39 SSP5-8.5 in the absence of Marine Ice Cliff Instability, and by up to 16 m under SSP5-8.5 considering 40 Marine Ice Cliff Instability (low confidence). {9.6.3} 41 42 Cryospheric Changes and Sea Level Rise at Specific Levels of Global Warming 43 44 At sustained warming levels between 1.5°C and 2°C, the Arctic Ocean will become practically sea ice– 45 free in September in some years (medium confidence); the ice sheets will continue to lose mass (high 46 confidence), but will not fully disintegrate on time scales of multiple centuries (medium confidence); there is 47 limited evidence that the Greenland and West Antarctic Ice Sheets will be lost almost completely and 48 irreversibly over multiple millennia; about 50–60% of current glacier mass excluding the two ice sheets and 49 the glaciers peripheral to the Antarctic Ice Sheet will remain, predominantly in the polar regions (low 50 confidence); Northern hemisphere spring snow cover extent will decrease by up to 20% relative to 1995– 51 2014 (medium confidence); the permafrost volume in the top 3 m will decrease by up to 50% relative to Do Not Cite, Quote or Distribute 9-9 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 1995–2014 (medium confidence). Committed GMSL rise over 2000 years will be about 2-6 m with 2°C of 2 peak warming (medium agreement, limited evidence). {9.3.1, 9.4.1, 9.4.2, 9.5.1, 9.5.2, 9.5.3, 9.6.3} 3 4 At sustained warming levels between 2°C and 3°C, the Arctic Ocean will be practically sea ice–free 5 throughout September in most years (medium confidence); there is limited evidence that the Greenland and 6 West Antarctic Ice Sheets will be lost almost completely and irreversibly over multiple millennia; both the 7 probability of their complete loss and the rate of mass loss will increase with higher temperatures (high 8 confidence); about 50–60% of current glacier mass outside Antarctica will be lost (low confidence); Northern 9 hemisphere spring snow cover extent will decrease by up to 30% relative to 1995–2014 (medium 10 confidence); permafrost volume in the top 3 m will decrease by up to 75% relative to 1995–2014 (medium 11 confidence). Committed GMSL rise over 2000 years will be about 4-10 m with 3°C of peak warming 12 (medium agreement, limited evidence). {9.3.1, 9.4.1, 9.4.2, 9.5.1, 9.5.2, 9.5.3, 9.6.3} 13 14 At sustained warming levels between 3°C and 5°C, the Arctic Ocean will become practically sea ice–free 15 throughout several months in most years (high confidence); near-complete loss of the Greenland Ice Sheet 16 and complete loss of the West Antarctic Ice Sheet will occur irreversibly over multiple millennia (medium 17 confidence); substantial parts or all of Wilkes Subglacial Basin in East Antarctica will be lost over multiple 18 millennia (low confidence); 60–75% of current glacier mass outside Antarctica will disappear (low 19 confidence); nearly all glacier mass in low latitudes, Central Europe, Caucasus, Western Canada and USA, 20 North Asia, Scandinavia and New Zealand will likely disappear; Northern Hemisphere spring snow cover 21 extent will decrease by up to 50% relative to 1995–2014 (medium confidence); permafrost volume in the top 22 3 m will decrease by up to 90% compared to 1995–2014 (medium confidence). Committed GMSL rise over 23 2000 years will be about 12–16 m with 4°C of peak warming and 19–22 m with 5°C of peak warming 24 (medium agreement, limited evidence). {9.3.1, 9.4.1, 9.4.2, 9.5.1, 9.5.2, 9.5.3, 9.6.3} 25 Do Not Cite, Quote or Distribute 9-10 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 9.1 Introduction 2 3 This chapter provides a holistic assessment of the physical processes underlying global and regional changes 4 in the ocean, cryosphere and sea level, as well as improved understanding of observed, attributed and 5 projected future changes since the AR5 and the SROCC (see outline in Figure 9.1). The ocean and 6 cryosphere (defined as the frozen components of the Earth system such as sea ice, ice sheets, glaciers, 7 permafrost and snow) exchange heat and freshwater with the atmosphere and each other (Figure 9.2). In a 8 warming climate, the combined effects of thermal expansion of seawater and melting of the terrestrial 9 cryosphere result in global mean sea-level rise (Box 9.1). 10 11 12 [START FIGURE 9.1 HERE] 13 14 Figure 9.1: Visual guide to chapter 9 with relevant chapter numbers indicated in red. 15 16 [END FIGURE 9.1 HERE] 17 18 19 Ocean acidification and deoxygenation are covered in Chapter 5 and regional changes to the ocean and 20 cryosphere are covered in Chapter 12 and the Atlas. Ecosystem range shifts and climate risk for marine 21 biodiversity associated with ocean change are assessed in AR6 Working Group II. The notion of “climate 22 velocity” often used in impact studies, which is defined as the speed and direction at which a climate 23 variable moves across a corresponding spatial field is underpinned by the assessment of changes in the 24 physical characteristics of the ocean provided in this chapter. 25 26 There are two major advances of this chapter compared with the AR5 and the SROCC facilitated by 27 community efforts. The first is the temporal and spatial increase in observations of both the ocean and the 28 cryosphere (Section 1.5.1.1). In particular, extended observations have allowed improved assessment of past 29 change and closure of both the energy and sea-level budget in a consistent way (Cross-Chapter Box 9.1) and 30 the sea level budget for the last century (Section 9.6.1.1). Higher resolution observations have revealed the 31 details of the Atlantic meridional overturning circulation (Section 9.2.3.1) and globally resolved glacier 32 changes for the first time (Section 9.5.1.1). Improved methodology has resulted in a doubling of the assessed 33 level of observed increase in global ocean 0–200 m stratification compared to the SROCC assessment 34 (Section 9.2.1.3). 35 36 The second advance is the use of a hierarchy of models and emulators to update projections of oceanic, 37 cryospheric and sea-level change arising from CMIP6 and related projects (Section 1.5.4.3; Table 1.3, Annex 38 II).2 CMIP6 included an ice sheet modelling intercomparison for the first time. Particular modelling 39 advances relevant to this chapter are the increase in ocean resolution in the HighResMIP and OMIP2 40 experiments (Section 1.5.3.1; Section 9.2), projections of future glacier (GlacierMIP) and ice sheet (ISMIP6 41 and LARMIP-2) response from multi-model studies (Sections 9.5.1, 9.4; Box 9.3), and new methods to 42 synthesize ocean and cryosphere models into sea level projections for all SSPs (Section 1.6.1; Cross-Chapter 43 Box 1.4; Sections 9.4.1.3, 9.4.2.5, 9.6.3) and warming levels (Sections 9.6.3; 1.6.2; Cross-Chapter Box 44 11.1). In particular, sea level projections and the individual contributions (Section 9.6.3.3) are consistent 45 with equilibrium climate sensitivity and surface temperature assessments across this report (Box 4.1, Cross- 46 Chapter Box 7.1). 47 48 There are other advances in scientific understanding. In the cryosphere, this chapter assesses how fast- 49 responding elements (sea ice, permafrost and snow (Sections 9.3 9.5.2; 9.5.3) track warming levels across 50 observations and projections independent of scenario, process understanding of uncertainty in Antarctic ice 2In particular, this range of tools leads to advance in evaluation of confidence in projections. When CMIP6 models are used without additional evidence, the 5-95% confidence range of projections is assigned to a likely range to acknowledge that there are uncertainty sources not reflected by model spread, consistent with Chapter 4. Do Not Cite, Quote or Distribute 9-11 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 sheet projections (Section 9.4.2; Box 9.4) and new insight into thresholds for Arctic sea ice (Section 9.3.1.1) 2 and Greenland and West Antarctic ice sheets (Section 9.4.1.4; 9.4.2.6). In the ocean, process understanding 3 of ocean heat uptake (Section 9.2.2.1; Cross-chapter Box 5.3) and observed changes in ocean stratification 4 (Section 9.2.1.3) have implications for ocean biogeochemistry are also important. 5 6 7 [START FIGURE 9.2 HERE] 8 9 Figure 9.2: Components of ocean, cryosphere and sea level assessed in this chapter. (a) Schematic of processes 10 (mCDW=modified Circumpolar Deep Water, GIA=Glacial Isostatic Adjustment). White arrows indicate 11 ocean circulation. Pinning points indicate where the grounding line is most stable and ice sheet retreat 12 will slow. (b) Geographic distribution of ocean and cryosphere components (numbers indicate (RGI 13 Consortium, 2017) glacierized regions, see Figures 9.20 and 9.21 for labels). Sea ice shaded to indicate 14 the annual mean concentration. Green ocean colours indicate larger surface current speed. Further details 15 on data sources and processing are available in the chapter data table (Table 9.SM.9). 16 17 [END FIGURE 9.2 HERE] 18 19 20 [START BOX 9.1 HERE] 21 22 BOX 9.1: Key processes driving sea level change 23 24 Sea-level change arises from processes acting on a range of spatial and temporal scales, in the ocean, 25 cryosphere, solid Earth, atmosphere and on land (Figure 9.2). Relative sea-level change is the change in 26 local mean sea surface height relative to the sea floor, as measured by instruments that are fixed to the 27 Earth’s surface (e.g., tide gauges). This reference frame is used when considering coastal impacts, hazards 28 and adaptation needs. In contrast, geocentric sea-level change is the change in local mean sea surface height 29 with respect to the terrestrial reference frame and is the sea-level change observed with instruments from 30 space. This box provides a brief summary of sea-level processes using standard terminology (Gregory et al., 31 2019). 32 33 Global processes 34 35 Global mean sea-level change (Sections 9.6, 2.3.3.3) is the change in volume of the ocean divided by the 36 ocean surface area. It is the sum of changes in ocean density (global mean thermosteric sea-level change) 37 and changes in the ocean mass as a result of changes in the cryosphere or land water storage (barystatic sea- 38 level change). 39 40 Steric sea-level change is caused by changes in the ocean density and is composed of thermosteric sea-level 41 change and halosteric sea-level change. Thermosteric sea-level change (also referred to as thermal 42 expansion) occurs as a result of changes in ocean temperature: increasing temperature reduces ocean density 43 and increases the volume per unit of mass. Halosteric sea-level change occurs as a result of salinity 44 variations: higher salinity leads to higher density and decreases the volume per unit of mass. Although both 45 processes can be relevant on regional to local scales, thermosteric changes contribute to global mean sea- 46 level change, whereas global mean halosteric change is negligible (Gregory et al., 2019). There is high 47 confidence in the understanding of processes causing thermosteric sea-level change (Section 9.2.4.1). 48 49 The Greenland and Antarctic ice sheets are the largest reservoirs of frozen freshwater and therefore 50 potentially the largest contributors to sea-level rise. Fluctuations in ice sheet volume arise from the 51 imbalance between accumulation (either at the ice sheet surface or on the underside of ice shelves) and loss 52 from sublimation, surface and basal melting, and iceberg calving. Ice sheets discharge the majority of their 53 mass through marine-terminating ice streams that are in some cases buttressed by floating ice shelves. 54 Changes in the thickness and extent of the ice shelves due to melt from below, calving, or disintegration, as a 55 result of surface meltwater penetrating crevasses, can affect the flow of the inland ice streams. There is 56 medium confidence in ice sheet processes but low confidence in their forcing (ocean changes and ice shelf Do Not Cite, Quote or Distribute 9-12 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 collapse) and in instability processes (Sections 9.4.1, 9.4.2).3 2 3 Glaciers contribute to sea-level change via an imbalance between mass gain and mass loss processes, which 4 leads to adjustments in the glacier geometry over an extended period of time, called the response time. The 5 response time may range from a few years to a few hundred years. The glacial meltwater does not all flow 6 immediately into the ocean: it can refreeze, feed rivers (where it may be extracted for domestic use), 7 evaporate, or be stored in (proglacial) lakes or closed basins. There is medium to high confidence in the 8 understanding of processes leading to sea level contributions from glaciers (Section 9.5.1). 9 10 Land water storage includes surface water, soil moisture, groundwater storage and snow, but excludes water 11 stored in glaciers and ice sheets. Changes in land water storage can be caused either by direct human 12 intervention in the water cycle (e.g., storage of water in reservoirs by building dams in rivers, groundwater 13 extraction for consumption and irrigation, or deforestation) or by climate variations (e.g., changes in the 14 amount of water in internally drained lakes and wetlands, the canopy, the soil, the permafrost and the 15 snowpack). Land water storage changes caused by climate variations may be indirectly affected by 16 anthropogenic influences. It is difficult to assign a single confidence level to land water storage as 17 understanding can vary from low confidence in groundwater recharge processes to high confidence in water 18 storage via snowpack changes (Sections 8.2.3, 8.3.1.7). 19 20 Regional and local processes 21 22 Ocean dynamic sea-level change refers to the change in mean sea level relative to the geoid and is 23 associated with the circulation and density-driven changes in the ocean. Ocean dynamic sea-level change 24 varies regionally but by definition has a zero global mean. It includes the depression of the sea surface by 25 atmospheric pressure. There is medium confidence in the understanding of ocean processes leading to 26 dynamic sea-level change (Section 9.2.4.2). 27 28 Changes in Earth Gravity, Earth Rotation and viscoelastic solid Earth Deformation (GRD) result from the 29 redistribution of mass between terrestrial ice and water reservoirs and the ocean. Contemporary terrestrial 30 mass loss leads to elastic solid Earth uplift and a nearby relative sea-level fall (for a single source of 31 terrestrial mass loss this is within ~2000 km, for multiple sources the distance depends on the interaction of 32 the different relative sea-level patterns). Farther away (more than ~7000 km for a single source of terrestrial 33 mass loss), relative sea level rises more than the global average, due, to first order, to gravitational effects. 34 Earth deformation associated with adding water to the ocean and a shift of the Earth’s rotation axis towards 35 the source of terrestrial mass loss leads to second-order effects that increase spatial variability of the pattern 36 globally. GRD effects due to the redistribution of ocean water within the ocean itself are referred to as self- 37 attraction and loading effects. There is high confidence in the understanding of GRD processes. 38 39 Glacial Isostatic Adjustment is ongoing GRD in response to past changes in the distribution of ice and water 40 on Earth’s surface. On a timescale of decades to tens of millennia following mass redistribution, Earth’s 41 mantle flows viscously as it evolves toward isostatic equilibrium, causing solid Earth movement and geoid 42 changes, which can result in regional to local sea-level variations. There is medium confidence in the 43 understanding of glacial isostatic adjustment processes. 44 45 Vertical land motion is the change in height of the land surface or the sea floor and can have several causes 46 in addition to elastic deformation associated with contemporary GRD and viscoelastic deformation 47 associated with glacial isostatic adjustment. Subsidence (sinking of the land surface or sea floor) can occur 48 through compaction of alluvial sediments in deltaic regions, removal of fluids such as gas, oil, and water, or 49 drainage of peatlands. Tectonic deformation of the Earth’s crust can occur as a result of earthquakes and 50 volcanic eruptions. There is medium confidence in the understanding of vertical land motion processes. 51 3The conversion of land ice mass loss to global mean sea-level rise used in this report (the Sea Level Equivalent, SLE) is 362.5 gigatons (Gt) of ice loss for 1 mm of sea-level rise Do Not Cite, Quote or Distribute 9-13 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Extreme sea level is an exceptionally low or high local sea-surface height arising from combined short-term 2 phenomena (e.g. storm surges, tides and waves). Relative sea-level changes affect extreme sea levels directly 3 by shifting the mean water levels and indirectly by modulating the depth for propagation of tides, waves 4 and/or surges. Extreme sea levels can be influenced by changes in the frequency, tracks, or strength of 5 weather systems, or anthropogenic changes such as dredging. Extreme Still Water Level refers to the 6 combined contribution of relative sea-level change, tides and storm surges. Wind-waves also contribute to 7 coastal sea level. Extreme Total Water Level is the extreme still water level plus wave setup (time-mean sea- 8 level elevation due to wave energy dissipation). When considering coastal impacts, swash (vertical 9 displacement up the shore-face induced by individual waves) is also important and included in Extreme 10 Coastal Water Level. There is low to medium confidence in the understanding of extreme sea level processes 11 (Sections 9.6.4, 12.4). 12 13 14 [END BOX 9.1 HERE] 15 16 17 9.2 Oceans 18 19 9.2.1 Ocean surface 20 21 9.2.1.1 Sea Surface Temperature (SST) 22 23 The AR5 (Hartmann et al., 2013) assessed that it is certain that global sea-surface temperature (SST) has 24 increased since the beginning of the 20th century (very high confidence). The SROCC did not assess past 25 SST change. Since the AR5, improvements in the understanding of recent SST biases in the observational 26 records, especially extending ship-based observations with buoy-based observations and improved treatment 27 of sea ice, have had important consequences for key climate change indicators such as global mean surface 28 temperature (GMST), global mean surface air temperature (GSAT), and SST (Cross-Chapter Box 2.3). The 29 AR5 assessment is confirmed, and it is now very likely that global mean SST changed by 0.88°C [0.68- 30 1.01°C] from 1850-1900 to 2011-2020, and 0.60°C [0.44-0.74°C] from 1980 to 2020 (Figure 9.3, Table 2.4). 31 32 Regions vary in the rate of SST warming, with slight cooling in some regions (Figure 9.3). The SROCC 33 (Collins et al., 2019) and Section 7.4.4 assess SST changes over specific regions, which are consistent with 34 the changes reported here. The tropical ocean has been warming faster than other regions since 1950, with 35 the fastest warming in regions of the tropical Indian and western Pacific Oceans (Figure 9.3), due to a 36 combination of local atmosphere-ocean coupling, the Indonesian Throughflow (Section 9.2.3.4; Figure 9.11), 37 and trends in the Walker circulation (Section 2.3.1.4.1; Section 3.3.3.1; Figure 3.16). The Western Boundary 38 Currents of the subtropical gyres have warmed faster than the global mean over the past century. There 39 remains low agreement in the changes of both the location and the dynamical changes in western boundary 40 current extensions (Sections 2.3.3.4.2, 9.2.3.4, Figure 9.3). In the Arctic, the mean SST increase over the last 41 two decades is similar to or only slightly higher than the global average (Chen et al., 2019b). In contrast, the 42 eastern Pacific Ocean, subpolar North Atlantic Ocean and Southern Ocean have warmed more slowly than 43 the global average or cooled. (Figure 9.3). Surface warming in the subpolar Southern Ocean has been slower 44 than the global average since the 1950s, and this pattern is consistent with the upwelling around Antarctica 45 renewing surface water with pre-industrial, deeper water-masses (Section 9.2.3.2) (Frölicher et al., 2015; 46 Marshall et al., 2015b; Armour et al., 2016). New evidence since the SROCC (Meredith et al., 2019) 47 confirms slight cooling since the 1980s around the subpolar Southern Ocean, contrasting with marked 48 warming directly northward of it (Section 9.2.3.2) (Haumann et al., 2020; Rye et al., 2020; Auger et al., 49 2021). In Eastern Boundary Upwelling Systems, the SROCC (Bindoff et al., 2019) reported low agreement 50 between SST trends in recent decades, due to varying spatio-temporal resolution and interannual to multi- 51 decadal variability. Satellite evidence not included in SROCC show that 92% of these regions warmed more 52 slowly than neighbouring offshore locations between 1982-2015, so upwelling may buffer the near shore 53 from warming (Varela et al., 2018) (Section 9.2.3.5). Coupled ocean-atmospheric modes of variability 54 strongly affect regional SST (Cross-Chapter Box 3.1, Annex IV). In summary, a positive SST trend since 55 1950 is evident globally, but there is very high confidence that the Indian Ocean, western equatorial Pacific Do Not Cite, Quote or Distribute 9-14 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Ocean, and western boundary currents have warmed faster than the global average, while the Southern 2 Ocean, the eastern equatorial Pacific, and the North Atlantic Ocean have warmed more slowly or slightly 3 cooled. 4 5 6 [START FIGURE 9.3 HERE] 7 8 Figure 9.3: Sea Surface Temperature (SST) and its changes with time. (a) Time series of global mean SST 9 anomaly relative to 1950-1980 climatology. Shown are observational reanalyses (HadISST) and multi- 10 model means from the CMIP historical, CMIP projections, and HighResMIP experiment. (b) Map of 11 observed SST (1995-2014 climatology HadISST), (e) bias of CMIP and (h) bias of HighResMIP (bottom 12 left) over 1995-2014. Also shown are 1950-2014 c) historical SST changes from observations, (f) CMIP 13 and (i) HighResMIP, and (d) 2005-2100 SST change rate, (g) 2005-2050 change rate for SSP5-8.5 for the 14 CMIP ensemble, and (j) 2005-2050 change rate for SSP5-8.5 for the HighResMIP ensemble. No overlay 15 indicates regions with high model agreement, where ≥80% of models agree on sign of change; diagonal 16 lines indicate regions with low model agreement, where <80% of models agree on sign of change (see 17 Cross-Chapter Box Atlas.1 for more information). Further details on data sources and processing are 18 available in the chapter data table (Table 9.SM.9). 19 20 [END FIGURE 9.3 HERE] 21 22 23 In the AR5 (Flato et al., 2013), a marginal improvement in CMIP5 climate model SST biases was noted 24 compared to CMIP3 models in the AR4, with a reduction in the magnitude of biases. The AR5 noted that in 25 several regions large SST biases are symptomatic of errors in the representation of important processes, such 26 as dynamics in the equatorial Pacific and North Atlantic, and Southern Ocean. Common regional biases in 27 SST or historical SST trends are not exclusively linked to the representation of the ocean (high confidence), 28 but can have multiple causes, including errors in the representation of long term historical trends in 29 equatorial winds (Section 9.2.1.2); misrepresentation of the forced equatorial ocean response (Karnauskas et 30 al., 2012; Kohyama et al., 2017; Coats and Karnauskas, 2018); thermocline depth errors (Linz et al., 2014); 31 errors in atmospheric model cloud-related short-wave radiation (Hyder et al., 2018); biases in ocean 32 circulation variability (Wang et al., 2014a); and deficiencies in upper ocean (Li et al., 2019b) and 33 atmospheric (Bates et al., 2012) boundary layer parameterizations. In CMIP6, the mid-latitude biases in the 34 northern hemisphere are improved in the multi-model mean and the inter-model standard deviation of the 35 zonal mean SST error is significantly decreased in the northern Hemisphere south of 50°N compared to 36 CMIP5, though biases in equatorial regions remains essentially unchanged (Section 3.5.1.1; Figures 3.23, 37 3.24, 9.3). Some longstanding ocean model biases have been reduced through increases in model resolution 38 in CMIP6 (Bock et al., 2020) and improved parameterizations (Fox-Kemper et al., 2011; Li et al., 2016a; 39 Qiao et al., 2016; Reichl and Hallberg, 2018). The high resolution HighResMIP ensemble (Figure 9.3) has 40 smaller cold biases in the North Atlantic and the tropical Pacific, and smaller warm biases in the upwelling 41 regions off the western coasts of Africa, North and South America (Roberts et al., 2018, 2019; Caldwell et 42 al., 2019; Docquier et al., 2019). In summary, CMIP6 models show persistent regional biases in representing 43 the climatological SST state (very high confidence), but higher resolution reduces some biases particularly in 44 the North Atlantic and Eastern Boundary Upwelling Systems (Figure 9.3; high confidence). 45 46 CMIP6 models represent the observed trends in SST patterns with greater fidelity than CMIP5, with the 47 ocean area that is inconsistent with the observed trends decreasing by about three quarters from CMIP5 to 48 CMIP6 (Olonscheck et al., 2020). In some regions, the direction of SST changes in observations are 49 consistent with CMIP6 only when including internal variability (Olonscheck et al., 2020). This is notably the 50 case in the equatorial Pacific, North Atlantic, and Southern Ocean, which are regions where SST is of known 51 importance in controlling heat uptake (Section 9.2.2.1) and the global radiative feedback parameter (Section 52 7.4.4.3). Overall, despite some persistent regional biases, CMIP6 coupled climate models reproduce the 53 observed SST trends or high internal variability over the past century over a range of different multidecadal 54 periods (Olonscheck et al., 2020; Watanabe et al., 2021) (Figure 9.3), highlighting their skill to inform future 55 large-scale SST changes at regional scale. Warming is projected at varying rates in all regions by 2050, 56 except the North Atlantic Subpolar Region, the equatorial Pacific, and the Southern Ocean where models Do Not Cite, Quote or Distribute 9-15 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 disagree (high confidence). 2 3 It is virtually certain that SST will continue to increase in the 21st century at a rate depending on future 4 emission scenario. The future global mean SST increase projected by CMIP6 models for the period 1995- 5 2014 to 2081-2100 is 0.86°C (5-95% range: 0.43-1.47°C) under SSP1-2.6, 1.51 °C [1.02-2.19°C] under 6 SSP2-4.5, 2.19°C (1.56-3.30°C) under SSP3-7.0, and 2.89°C (2.01-4.07°C) under SSP5-8.5 (Figure 9.3). 7 While under SSP1-2.6, the CMIP6 ensemble consistently projects that it is very likely at least 83% of the 8 world ocean surface will have warmed by 2100, under SSP5-8.5, at least 98% of the world ocean surface 9 will have warmed. The spatial pattern of future change is consistent with observed SST change over the 20 th 10 century, though with notable regional differences (Figure 9.3). Long-term change in SST patterns is 11 important for regional impacts but also affects radiative feedbacks, and therefore long-term change in 12 climate sensitivity (Section 7.4.4.3). In the Southern Ocean, CMIP6 models project that SSTs will eventually 13 consistently increase in the 21st century at a rate dependent on future scenario (Bracegirdle et al., 2020) 14 (Figure 9.3, Section 9.2.3.2). Yet, there is only low confidence that this Southern Ocean warming will 15 emerge by the end of the century (Section 7.4.4.1), due the inconsistent historical and near-term simulations 16 and observations over the 20th century (Figure 9.3). Furthermore, the equilibrium SST pattern from proxy 17 records or simulated by climate models under CO2 forcing stand in contrast with the cooling trends in the 18 Southern Ocean observed over the past decades (Section 7.4.4.1.2). Similarly, the SST change pattern 19 observed in the tropical Pacific Ocean will transition on centennial timescales to a mean pattern resembling 20 the El Niño pattern (medium confidence, Annex IV). However, it is difficult to delineate a climate change 21 trend ressembling an El Niño pattern and El Niño variability (Wittenberg, 2009; Collins et al., 2010) without 22 large ensembles (Kay et al., 2015). Several Pliocene SST reconstructions indicate enhanced warming in the 23 centre of the eastern Pacific equatorial cold tongue upwelling region, consistent with reconstruction of 24 enhanced subsurface warming and enhanced warming in coastal upwelling regions (Section 7.4.4.2.2). The 25 North Atlantic subpolar gyre is projected to continue to warm more slowly than surrounding regions (Suo et 26 al., 2017), as the Gulf Stream concurrently warms rapidly (Cheng et al., 2013) (Figure 9.3) and the Atlantic 27 Meridional Overturning Circulation further declines under greenhouse gas forcing although models disagree 28 about the rate of change (Figure 9.3, Section 9.2.3.1). In summary, CMIP6 models show a future pattern of 29 SST change comparable to historical trends with intensity depending on future emission scenario, and some 30 of the observed cooling trends over the 20th century will eventually transition to a warming SST on 31 centennial timescales, in particular in the Southern Ocean (high confidence) and in the equatorial Pacific 32 (medium confidence), while the North Atlantic subpolar gyre will continue to warm more slowly than the 33 global average (high confidence). 34 35 36 9.2.1.2 Air-sea fluxes 37 38 Air-sea fluxes of energy, freshwater, and momentum (wind stresses) are difficult to observe directly (Cronin 39 et al., 2019), so estimates of the global mean net air-sea heat flux are inferred from observed ocean warming 40 (Section 2.3.3.1, Box 7.2, and Cross-Chapter Box 9.1). Air-Sea heat fluxes resemble the warming patterns of 41 CMIP3 (Domingues et al., 2008; Levitus et al., 2012) and are consistent with the ensemble mean warming 42 rate of CMIP5 (Cheng et al., 2017, 2019) and CMIP6 models (Section 3.5.1.3). Regional air-sea fluxes in 43 models remain a key driver of uncertainty (Huber and Zanna, 2017; Tsujino et al., 2020). A substantial part 44 of the upper 700 m energy increase is very likely attributed to anthropogenic forcing via increasing radiative 45 forcing (Sections 3.5.1.3, 7.2, 7.3). 46 47 The SROCC (Abram et al., 2019) and AR5 (Rhein et al., 2013) assessed that observations of air-sea fluxes 48 had not yet reached the density or accuracy to directly detect trends beyond the noise. New evidence since 49 SROCC confirms that direct heat and freshwater flux trends have not emerged yet as spatial (Figure 9.4), 50 annual (Yu, 2019), and decadal (Zanna et al., 2019a) variability overwhelm detection. Since the AR5, 51 comprehensive comparisons (Bentamy et al., 2017; Valdivieso et al., 2017; Yu et al., 2017) have used 52 updated and new surface flux products to improve surface flux uncertainty estimates, and these comparisons 53 note that implied global energy imbalances often exceed the observed ocean warming. Flux estimates using 54 top of atmosphere observations and atmospheric fluxes from reanalysis have improved over past products 55 (Trenberth and Fasullo, 2018) but require consistency adjustments (Trenberth et al., 2019) as the energy Do Not Cite, Quote or Distribute 9-16 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 budget is not closed. Adjustments are needed for all flux products, and they remain less accurate than direct 2 ocean heat content change measurements (Cheng et al., 2017). Some regional changes are likely robust in 3 both satellite observations and projections (Figure 9.4). Recent satellite-based surface flux products with 4 improved retrieval algorithms and new satellites, e.g., J-OFURO3 (Tomita et al., 2019) and OAFlux-HR 5 (Yu, 2019), provide a complete suite of turbulent fluxes including heat, moisture, and momentum. When 6 combined with satellite-based surface radiation from CERES EBAF (Kato et al., 2018) and precipitation 7 from GPCP (Adler et al., 2003), full ocean-surface forcing is available since 1987 (Figure 9.4). These 8 products agree with sparse buoy and ship observations within 30 W m-2 (Bentamy et al., 2017; Cronin et al., 9 2019). While patterns agree between models and satellites in net fluxes (Figure 9.4), the trend magnitudes 10 are substantially weaker in models. The fluxes tending to warm the North Atlantic and Southern Ocean are 11 consistent with the largest changes observed in the surface properties and water-masses (Sections 9.2.1.1, 12 9.2.2.1, 9.2.2.3). The observed trend toward a saltier Atlantic Ocean and a fresher Indian Ocean as well as 13 trends in evaporation minus precipitation (E-P) patterns in the equatorial Pacific (see also Section 8.3.1) 14 enhance the present mean pattern of wetting and drying. Elsewhere patterns are less clear with only partial, 15 large-scale agreement with the wet gets wetter simplification (Sections 3.3.2.2, 4.4.1, 4.5.1). In summary, 16 globally-integrated and large-scale fluxes are more reliably inferred from heat content and salinity change, 17 while regional trends are rarely robust in observations and where robust tend to be underestimated or in 18 disagreement in models (very high confidence). 19 20 21 [START FIGURE 9.4 HERE] 22 23 Figure 9.4: Global maps of observed mean fluxes (a, d, g), the observed trends in these fluxes (b, e, h) and the 24 projected rate of change in these fluxes from SSP5-8.5 (c, f, i). Shown are the freshwater flux (a, b, c), 25 net heat flux (d, e, f), and momentum flux or wind stress magnitude (g, h, i), with positive numbers 26 indicating ocean freshening, warming, and accelerating respectively. The means and observed trends are 27 calculated between 1995-2014 (freshwater and wind stress) or 2001-2014 (heat) and the SSP5-8.5 28 projected rates are between 1995-2100 using 20-year averages at each end of the time period. 29 Observations show objective interpolation from CERES EBAF v4 (Kato et al., 2018), OAFlux-HR (Yu, 30 2019), and GPCP (Adler et al., 2003) of fluxes and flux trends. (b, e, h) Observed trends with no overlay 31 indicates regions where the trends are significant at p = 0.34 level. Crosses indicate regions where trends 32 are not significant. For (c, f, i) projections, no overlay indicates regions with high model agreement, 33 where ≥80% of models agree on the sign of change; diagonal lines indicate regions with low model 34 agreement, where <80% of models agree on the sign of change (see Cross-Chapter Box Atlas.1 for more 35 information). Further details on data sources and processing are available in the chapter data table (Table 36 9.SM.9). 37 38 [END FIGURE 9.4 HERE] 39 40 41 There is low confidence in long-term wind stress trends in most regions, but a few locations have likely 42 trends over the scatterometer era and in projections, as shown in Figure 9.4 (Desbiolles et al., 2017; Young 43 and Ribal, 2019; Yu, 2019). The AR5 (Rhein et al., 2013) assessed with medium confidence that zonal wind 44 stress over the Southern Ocean increased from the early 1980s to the 1990s (Figure 9.4, medium confidence). 45 Over 1995-2014, the zonal wind stress over the Southern Ocean continued to increase, westerly winds in the 46 North Pacific and North Atlantic weakened while the easterly equatorial Pacific winds of the Walker 47 circulation strengthened (Figure 9.4). In historical simulations, CMIP5 models projected annular modes 48 (Annex IV) to move poleward and strengthen in both hemispheres (Yang et al., 2016a), while in CMIP6 49 models westerlies only strengthen over the Southern Ocean, with a weaker trend than recently observed 50 (Figure 9.4, Sections 4.5.1, 4.5.3). In the tropical Pacific Ocean, a weakening trend in easterly winds and 51 Walker circulation in the 20th century has been inferred based on observed sea level pressure data (Vecchi et 52 al., 2006; Vecchi and Soden, 2007) and coral proxies (Carilli et al., 2014) and is projected to continue by 53 CMIP6 models (Figure 9.4), yet over 1995-2014 observed winds have strengthened (Figure 9.4). The 54 observed strengthening may have been influenced by a combination of factors (Section 7.4.4.2.1), but there 55 is low confidence in the attribution of this signal to anthropogenic warming (Section 3.3.3.1) and medium 56 confidence that it reflects internal variability (Section 8.3.2.3). Near-term projected changes over the Do Not Cite, Quote or Distribute 9-17 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Southern Ocean result from ozone recovery and greenhouse gases (Sections 4.3.3, 4.4.3). Overall, there is 2 only low confidence in observed and projected wind stress trends in most regions because trends in oceanic 3 wind stresses during the satellite era have not emerged or are inconsistent with historical simulated changes. 4 5 Air-sea flux biases result from common causes in most models, many are the same as during AR5 (Rhein et 6 al., 2013). Important currents (e.g., Gulf Stream, Kuroshio, Antarctic Circum-polar Current patterns) are 7 often found in erroneous locations in models, affecting SST and flux signatures (Bates et al., 2012; Beadling 8 et al., 2020; Li et al., 2020b), but their locations are improved in high-resolution ocean models (Chassignet et 9 al., 2017, 2020; Hewitt et al., 2020), and high-resolution coupled models reduce the mean air-sea flux biases 10 (Delworth et al., 2012; Sakamoto et al., 2012; Small et al., 2014a; Haarsma et al., 2016; Caldwell et al., 11 2019; Jackson et al., 2020a). Oceanic variability stems either from internal chaotic variability or atmospheric 12 forcing (Hasselmann, 1976; Sérazin et al., 2016, 2017). Large scale variability in the ocean tends to follow 13 atmospheric forcing in low resolution models, while in high-resolution coupled models ocean variability 14 drives atmospheric variability on small scales (Bishop et al., 2017; Small et al., 2019), allowing these high- 15 resolution models to mimic the coupling with clouds, precipitation, and atmospheric and oceanic boundary 16 layers apparent in observations (Chelton and Xie, 2010; Frenger et al., 2013). Even coarse resolution models, 17 such as the ocean and sea-ice components used in CMIP6, show significant sensitivity in the mean and 18 variability of SST and sea ice to modest changes in flux forcing (Tsujino et al., 2020). Finally, there is still 19 considerable disagreement between different parameterizations of air-sea fluxes used in models and strong 20 scatter in direct observations (Renault et al., 2016; Brodeau et al., 2017). In summary, there is very high 21 confidence that air-sea heat flux and stress biases are reduced in coupled models with high ocean resolution 22 over coarse resolution models, although the effect on trends remain unclear. 23 24 25 9.2.1.3 Upper Ocean Stratification and Surface Mixed Layers 26 27 The density difference from surface to deep ocean is the upper-ocean stratification. The AR5 (Rhein et al., 28 2013) assessed that it is very likely that the thermal contribution to stratification over the fixed 0-200 m layer 29 increased by about 1% per decade between 1971 and 2010 (based on linear trend consistently across reports). 30 The SROCC (Bindoff et al., 2019) found it very likely that density stratification increased by 0.46-0.51% per 31 decade between 60°S and 60°N from 1970 to 2017). New published estimates based on a variety of different 32 interpolated observations show that the SROCC assessed rate is too low, even using the same data and 33 methods (Li et al., 2020a). The 1960-2018 stratification increase is estimated at 1.2±0.1% per decade from 34 the IAP product, 1.2±0.4% per decade from the Ishii product, 0.7±0.5% per decade from the EN4 product, 35 0.9±0.5% per decade from the ORAS4 product, and 1.2 ±0.3% per decade from the NCEI product (Li et al., 36 2020a). The improved methodology of computing stratification change on individual profiles before gridding 37 yields a global annual mean increase of 0-200 m stratification change of 0.8±0.2% per decade between 1960 38 and 2018 (Yamaguchi and Suga, 2019) and a global summer mean increase of 0-200 m stratification change 39 of 1.3±0.3% per decade between 1970 and 2018 (Sallée et al., 2021)is of a similar magnitude to the long 40 term trend (Yamaguchi and Suga, 2019; Li et al., 2020a). In summary, there is limited evidence that focusing 41 on changes over a fixed depth range might hide larger increases occurring at the seasonally and regionally 42 variable pycnocline depth. There is also limited evidence that summer stratification change within the 43 pycnocline has occurred at a rate of 8.9±2.7% per decade from 1970 to 2018, and limited evidence of a 44 winter pycnocline stratification increase (Cummins and Ross, 2020a; Sallée et al., 2021). 45 46 47 48 [START FIGURE 9.5 HERE] 49 50 Figure 9.5: Mixed layer depth in (a-d) winter and (e-h) summer. (a, e) Observed climatological mean mixed layer 51 depth (based on density threshold) from the Argo Mixed Layer Depth (Holte et al., 2017) observations 52 2000-2019. (b, f) Bias between the observation-based estimate (2000-2019) and the 1995-2014 CMIP6 53 climatological mean mixed layer depth. (c, d, g, h) Projected MLD change from 1995-2014 to 2081-2100 54 under (c, g) SSP1-2.6 and (d, h) SSP5-8.5 scenarios. The (a, b, c, d) Winter row shows DJF in the 55 Northern Hemisphere and JJA in the Southern Hemisphere, and the (e, f, g, h) Summer row shows JJA in 56 the Northern Hemisphere and DJF in the Southern Hemisphere. The mixed layer depth is the depth where Do Not Cite, Quote or Distribute 9-18 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 the potential density is 0.03 kg m-3 denser than at 10m. No overlay indicates regions with high model 2 agreement, where ≥80% of models agree on the sign of change; diagonal lines indicate regions with low 3 model agreement, where <80% of models agree on the sign of change (see Cross-Chapter Box Atlas.1 for 4 more information). Further details on data sources and processing are available in the chapter data table 5 (Table 9.SM.9). 6 7 [END FIGURE 9.5 HERE] 8 9 10 While AR5 and the SROCC did not assess change in mixed-layer depth, the reported changes in 11 stratification can modulate the surface mixed layer depth, which is set by a balance between fluxes and 12 dynamical mixing (winds, tides, waves, convection) acting against the background stratification and 13 restratification processes (solar and dynamical). Despite the large stratification increase observed at a global 14 scale, new evidence shows that summer mixed-layer depth deepened consistently over the globe at a rate of 15 2.9±0.5% per decade from 1970 to 2018, with the largest deepening observed in the Southern Ocean, 16 corresponding to overall deepening from 3-15 m per decade depending on region (Somavilla et al., 2017; 17 Sallée et al., 2021). While the shorter observational record in winter than in summer does not allow global 18 winter mixed-layer trends to be reliably assessed (Sallée et al., 2021), winter mixed-layer depths deepening 19 at rates of 10 m per decade have been reported at individual long-term mid-latitude monitoring sites 20 (Somavilla et al., 2017). Projections agree that shoaling of mixed layer depth is expected in the 21st century, 21 but only for strong emissions scenarios and only in some regions (Figure 9.5). In summary, there is limited 22 observational evidence that the mixed layer is globally deepening, while models show no emergence of a 23 trend until later in the 21st century under strong emissions. 24 25 The SROCC assessed that upper ocean stratification will continue to increase in the 21st century under 26 increased radiative forcing (high confidence), due to increased surface temperature and high-latitude surface 27 freshening (Bindoff et al., 2019). New climate model simulations concur with the SROCC assessment of a 28 future increase of the 0-200 m stratification under increased radiative forcing in all regions of the world 29 ocean (Kwiatkowski et al., 2020). In addition, CMIP6 climate models project a shallowing of the mixed- 30 layer both in summer and winter by the end of the century under increased radiative forcing (Kwiatkowski et 31 al., 2020) (Figure 9.5), with the exception of the Arctic showing deepening of the mixed-layer as a result of 32 sea-ice retreat (Lique et al., 2018a) (Figure 9.5). The regions of largest shallowing are associated with the 33 deepest climatological mixed layer both in winter and summer, particularly affecting the North Atlantic and 34 the Southern Ocean basins (Figure 9.5). While CMIP6 models tend to project shallowing mixed-layers under 35 a warming climate, except at high latitudes (Figure 9.5) (Lique et al., 2018b; Kwiatkowski et al., 2020), a 36 deepening in the summer mixed-layer depth by intensification of the surface winds and storms may explain 37 inconsistency among models in many regions (Figure 9.5) (Young and Ribal, 2019), although model mixed 38 layer biases are large in the summertime in the Southern Ocean (Belcher et al., 2012; Sallée et al., 2013a; Li 39 et al., 2016a; Tsujino et al., 2020). Lack of observed ocean turbulence and climate model limitations do not 40 allow for direct assessment of ocean surface turbulence change and limit confidence in past and future 41 mixed-layer change. Understanding of turbulent processes, its representation in ocean and climate models, 42 and its effect on mixed layer biases have been an active and rapidly evolving topic of research since AR5 43 (Buckingham et al., 2019; Li et al., 2019b). Small-scale mixed layer processes are not resolved in climate 44 models (D’Asaro, 2014; Buckingham et al., 2019; McWilliams, 2019) and despite significant improvements 45 in their parameterisation over the last decade (Fox-Kemper et al., 2011; Jochum et al., 2013; Li et al., 2016a; 46 Qiao et al., 2016; Li et al., 2019b) and significant improvement in some models (Li and Fox-Kemper, 2017; 47 Dunne et al., 2020), biases in mixed-layer representation generally persist (Heuzé, 2017; Williams et al., 48 2018; Cherchi et al., 2019; Golaz et al., 2019; Voldoire et al., 2019; Yukimoto et al., 2019; Boucher et al., 49 2020; Danabasoglu et al., 2020; Dunne et al., 2020; Kelley et al., 2020). In summary, the representation of 50 upper ocean stratification and mixed layers has improved in CMIP6 compared to CMIP5. While it is 51 virtually certain that the global mean upper ocean will continue to stratify in the 21st century, there is only 52 low confidence in the future evolution of mixed-layer depth, which is projected to mostly shoal under high 53 emissions scenarios except in high latitude regions where sea-ice retreats. 54 55 Do Not Cite, Quote or Distribute 9-19 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 [START BOX 9.2 HERE] 2 3 BOX 9.2: Marine Heatwaves 4 5 Marine heatwaves (MHW) are periods of extreme high sea temperature relative to the long-term mean 6 seasonal cycle (Hobday et al., 2016). Studies since the SROCC (Collins et al., 2019) confirm the assessment 7 that MHW can lead to severe and persistent impacts on marine ecosystems, from mass mortality of benthic 8 communities including coral bleaching, changes in phytoplankton blooms, shifts in species composition and 9 geographical distribution, toxic algal blooms to decline in fisheries catch and mariculture (Smale et al., 2019; 10 Cheung and Frölicher, 2020; Hayashida et al., 2020a; Piatt et al., 2020). Unlike synoptic atmospheric 11 heatwaves (Section 11.3), MHWs can extend for millions of square kilometres, persist for weeks to months, 12 and occur at subsurface (Bond et al., 2015; Schaeffer and Roughan, 2017; Perkins-Kirkpatrick et al., 2018; 13 Laufkötter et al., 2020). 14 15 The SROCC established that MHWs have occurred in all basins over the last decades. Additional evidence 16 documenting widespread occurrence of marine heat waves in all basins and marginal seas continues to 17 accumulate (Li et al., 2019c; Yao et al., 2020). The SROCC highlighted the role of large-scale climate modes 18 of variability in amplifying or suppressing MHW occurrences, which has since been further corroborated, 19 increasing confidence in climate modes as important drivers of MHWs (Holbrook et al., 2019; Sen Gupta et 20 al., 2020). More generally, understanding of processes leading to MHWs has increased since the SROCC, 21 including air-sea heat flux (Section 9.2.1.2), increased horizontal heat advection, shoaling of the mixed layer 22 and suppressed mixing processes (Section 9.2.1.3), reduced coastal upwelling and Ekman pumping (Section 23 9.2.3.5), changes in eddy activities and planetary waves, and the re-emergence of warm subsurface 24 anomalies (Holbrook et al., 2020; Sen Gupta et al., 2020). 25 26 The SROCC reported with high confidence that MHWs (defined as days exceeding the 99th percentile in SST 27 from 1982 to 2016) have very likely doubled in frequency between 1982 and 2016. Additional observation- 28 based evidence and acquisition of longer observation time-series since the SROCC have confirmed and 29 expanded on this assessment: since the 1980s MHWs have also become more intense and longer (Frölicher 30 and Laufkötter, 2018; Smale et al., 2019; Laufkötter et al. 2020). Satellite observations and reanalyses of 31 SST show increase in intensity of 0.04°C per decade from 1982 to 2016, an increase in spatial extent of 19 % 32 per decade from 1982 to 2016, and an increase in annual MHW days of 54 % between the 1987-2016 period 33 compared to 1925-1954 (Frölicher et al., 2018; Oliver, 2019). The SROCC assessed that 84-90% of all 34 MHWs that occurred between 2006 and 2015 are very likely caused by anthropogenic warming. There is new 35 evidence since SROCC that the frequency of the most impactful marine heatwaves over the last few decades 36 has increased more than 20-fold because of anthropogenic global warming (Laufkötter et al., 2020). In 37 summary, there is high confidence that MHWs have increased in frequency over the 20th century, with an 38 approximate doubling from 1982-2016, and medium confidence that they have become more intense and 39 longer since the 1980s. 40 41 42 [START BOX 9.2, FIGURE 1 HERE] 43 44 Box 9.2, Figure 1: Observed and simulated regional probability ratio of marine heatwaves (MHWs) for the 1985- 45 2014 period and for the end of the 21st century under two different greenhouse gas emissions 46 scenarios. The probability ratio is the proportion by which the number of MHW days per year has 47 increased relative to pre-industrial times. A MHW is defined as a deviation beyond the daily 99th 48 percentile (11-day window) in the deseasonalized sea surface temperature. (a) The MHW 49 probability ratio from satellite observations (NOAA OISST V2.1; (Huang et al., 2021) during 1985- 50 2014. The mean warming pattern (difference in ERSST5 (Huang et al., 2017) sea surface 51 temperature between the 1985-2014 and 1854-1900 periods) has been added to the satellite 52 observations to calculate the probability ratio. (b-d) CMIP6 simulated multi-model mean probability 53 ratio of the (b) 1985-2014 period, and 2081-2100 period in the (c) SSP1 2.6 and (d) SSP5 8.5 54 scenarios. The areas with grey diagonal lines in (d) indicate permanent MHWs (> 360 heatwave 55 days per year). These 14 CMIP6 models are included in the analysis: ACCESS-CM2, CESM2, 56 CESM2-WACCM, CMCCCM2-SR5, CNRM-CM6-1, CNRM-ESM2-1, CanESM5, EC-Earth3, Do Not Cite, Quote or Distribute 9-20 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 IPSL-CM6A-LR, MIROC6, MRI-ESM2-0, NESM3, NorESM2-LM, NorESM2-MM. Further 2 details on data sources and processing are available in the chapter data table (Table 9.SM.9). 3 4 [END BOX 9.2, FIGURE 1 HERE] 5 6 7 Consistent with the SROCC, future MHWs are defined with reference to the historical climate conditions. 8 The SROCC assessed that MHWs will very likely further increase in frequency, duration, spatial extent and 9 intensity under future global warming in the 21st century. CMIP6 projections allow us to confirm this 10 assessment and quantify future change based on global mean probability ratio change (Box 9.2, Figure 1): 11 they project MHWs will become 4 (5-95% range: 2-9) times more frequent in 2081-2100 compared to 1995- 12 2014 under SSP1-2.6, or 8 (3-15) times more frequent under SSP5-8.5. The SROCC highlighted that future 13 change of MHWs will not be globally uniform, with the largest changes in the frequency of marine 14 heatwaves being projected to occur in the western tropical Pacific and the Arctic Ocean (medium 15 confidence). New evidence from the latest generation of climate models confirms and complements the 16 SROCC assessment (Box 9.2, Figure 1). Moderate increases are projected for mid-latitudes, and only small 17 increases are projected for the Southern Ocean (medium confidence) (Hayashida et al., 2020b). While under 18 the SSP5-8.5 scenario, permanent MHWs (more than 360 days per year) are projected to occur in the 21st 19 century in parts of the tropical ocean, the Arctic Ocean and around 45°S, the occurrence of such permanent 20 MHWs can largely be avoided under the SSP1-2.6 scenario (Frölicher et al., 2018; Oliver et al., 2019; Plecha 21 and Soares, 2020). The resolution of current climate models (CMIP5 and CMIP6) capture the broad features 22 of MHWs, but they may have a bias towards weaker and longer MHWs in the historical period (medium 23 confidence) (Frölicher et al., 2018; Pilo et al., 2019; Plecha and Soares, 2020) and greater intensification in 24 western boundary current regions (Hayashida et al., 2020b). 25 26 27 [END BOX 9.2 HERE] 28 29 30 9.2.2 Changes in Heat and Salinity 31 32 9.2.2.1 Ocean Heat Content and Heat Transport 33 34 Ocean warming, i.e., changing ocean heat content, is an important aspect of energy on Earth: the SROCC 35 (Bindoff et al., 2019) reported that there is high confidence that ocean warming during 1971-2010 dominated 36 the increase in the Earth’s energy inventory, which is confirmed by the Box 7.2 assessment that the ocean 37 has stored 91% of the total energy gained from 1971 to 2018. As reported in Sections 2.3.3.1, 3.5.1.3 and 38 7.2.2.2, Box 7.2 and Cross-Chapter Box 9.1, confidence in the assessment of global ocean heat content 39 (OHC) change since 1971 is strengthened compared to previous reports and extended backward to include 40 likely warming since 1871. Table 7.1 updates the estimates of total ocean heat gains from 1971 to 2018, 1993 41 to 2018 and 2006 to 2018. Section 3.5.1.3 assesses that it is extremely likely that anthropogenic forcing was 42 the main driver of the OHC increase over the historical period. Section 2.3.3.1 reports that current 43 multidecadal to centennial rates of OHC gain are greater than at any point since the last deglaciation 44 (medium confidence). 45 46 Ocean warming is not uniform with depth. The AR5 (Rhein et al., 2013) assessed that since 1971 ocean 47 warming was virtually certain for the upper 700 m and likely for the 700-2000 m layer. Both the AR5 and 48 the SROCC (Bindoff et al., 2019) assessed that the deep ocean below 2000 m had likely warmed since 1992, 49 especially in the Southern Ocean. Section 2.3.3.1 provides an updated assessment of ocean temperature 50 change for different depth layers, different time periods and different observation-based reconstructions 51 (Table 2.7). Section 2.3.3.1 confirms the previous assessment that it is virtually certain that the upper ocean 52 (0-700m) has warmed since 1971, that ocean warming at intermediate depths (700-2000 m) is very likely 53 since 2006, and that it is likely that ocean warming has occurred below 2000 m since 1992. Section 3.5.1.3 54 assessed that it is extremely likely that human influence was the main driver of the ocean heat content 55 increase observed since the 1970s, which extends into the deeper ocean (very high confidence), and shows Do Not Cite, Quote or Distribute 9-21 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 that biases in potential temperature have a complex pattern (Figure 3.25). In the present section, we assess 2 the regional patterns of this warming and associated processes driving regional ocean warming. 3 4 The rate of ocean warming varies regionally, with some regions having experienced slight cooling (Figure 5 9.6). The SROCC (Bindoff et al., 2019) assessed that ocean warming in the 0-700 m is globally widespread, 6 with slower than global average warming in the subpolar North Atlantic. The SROCC (Meredith et al., 2019) 7 also estimated that the Southern Ocean accounted for ~75% of global ocean heat uptake during 1870–1995 8 and that 35-43% of the upper 2000 m global ocean warming occurred in the Southern Ocean over 1970-2017 9 (45-62% for 2005-2017). The SROCC noted that this interhemispheric asymmetry might at least partially be 10 explained by high concentrations of aerosols in the northern hemisphere. Here, we confirm these 11 assessments, bring new evidence attributing these regional trends, and discuss the role of decadal ocean 12 circulation variability in redistributing heat, driving interhemispheric asymmetry of the recent rate of ocean 13 warming (Rathore et al., 2020; Wang et al., 2021b). Since the SROCC, one new study shows that the 14 subpolar North Atlantic “warming hole” observed since the 1980s has emerged from internal climate 15 variability and can be attributed to greenhouse gas emissions (Chemke et al., 2020). A new analysis of a 16 suite of climate models (Hobbs et al., 2020) confirms the SROCC assessment, based on one paper (Swart et 17 al., 2018), attributing the observed Southern Ocean warming to anthropogenic forcing. Given the large 18 fraction of global ocean warming in the Southern Ocean and the sparse observations there before 2005, there 19 is limited evidence that global OHC increase since 1971 might have been underestimated (Cheng and Zhu, 20 2014; Durack et al., 2014). Cross-Chapter Box 9.1 accounts for an increased error before 2005 in global 21 OHC change. In summary, in the upper 2000 m since the 1970s, the subpolar North Atlantic has been slowly 22 warming, and the Southern Ocean has stored a disproportionally large amount of anthropogenic heat 23 (medium confidence). 24 25 26 [START FIGURE 9.6 HERE] 27 28 Figure 9.6: Ocean heat content (OHC) and its changes with time. (a) Time series of global ocean heat content 29 anomaly relative to a 2005-2014 climatology in the upper 2000m of the ocean. Shown are observations 30 (Ishii et al., 2017; Baggenstos et al., 2019; Shackleton et al., 2020), model-observation hybrids (Cheng et 31 al., 2019; Zanna et al., 2019a), and multi-model means from the CMIP6 historical (29 models) and SSP 32 scenarios (label subscripts indicate number of models per SSP). (b-g) Maps of Ocean Heat Content across 33 different time periods, in different layers, and from different data sets/experiments. Maps show the 34 CMIP6 ensemble bias and observed (Ishii et al., 2017) trends of OHC for (b, c) 0-700m for the period 35 1971-2014, and (e, f) 0-2000m for the period 2005-2017. CMIP6 ensemble mean maps show projected 36 rate of change 2015-2100 for (d) SSP5-8.5 and (g) SSP1-2.6 scenarios. Also shown are the projected 37 changes in 0-700m OHC for (d) SSP1-2.6 and (g) SSP5-8.5 in the CMIP6 ensembles, for the period 38 2091-2100 versus 2005-2014. No overlay indicates regions with high model agreement, where ≥80% of 39 models agree on the sign of change; diagonal lines indicate regions with low model agreement, where 40 <80% of models agree on the sign of change (see Cross-Chapter Box Atlas.1 for more information). 41 Further details on data sources and processing are available in the chapter data table (Table 9.SM.9). 42 43 [END FIGURE 9.6 HERE] 44 45 46 Below 2000 m, direct observations of full-depth ocean temperature change are limited to ship-based, high- 47 quality deep ocean temperature measurements. Such high quality full-depth ship-based sampling has 48 improved from 1990 to the present due to the World Ocean Circulation Experiment (WOCE) and the Global 49 Ocean Ship-based Hydrographic Investigations Program (GO-SHIP) (Sloyan et al., 2019). The SROCC 50 (Bindoff et al., 2019) assessed that the likely warming of the ocean since the 1990s below 2000 m is 51 associated with a marked regional pattern, with larger warming in the Southern Ocean. In the deep North 52 Atlantic, warming has reversed to cooling over the past decade, possibly due to internal variability fed by 53 North Atlantic Deep Water (Section 9.2.2.3). Over the past decade, the warming rate of Antarctic Bottom 54 Water (AABW, Section 9.2.2.3) has been dependent on origin: slower from the Weddell Sea and faster from 55 the Ross Sea and Adélie Land. One new study (Purkey et al., 2019a) strengthens confidence in AABW 56 warming: below 4000 m a monotonic, basin‐wide, and multidecadal temperature change is found in the Do Not Cite, Quote or Distribute 9-22 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 southern Pacific basin, with larger warming rates near the bottom water formation sites than further 2 downstream. New analysis of one model provides limited evidence that the sparse observational record may 3 underestimate the rate of deep ocean warming over 1990-2010 by about 20% (Garry et al., 2019) which is 4 included in the assessed OHC error (Cross-Chapter Box 9.1). There is still low agreement in deep ocean 5 changes from ocean data-assimilation reanalyses (Palmer et al., 2017) and low confidence in such inferences. 6 In summary, while observational coverage below 2000 m is sparser than in the upper 2000 m, there is high 7 confidence that deep ocean warming below 2000 m has been larger in the Southern Ocean than in other 8 ocean basins due to widespread AABW warming. 9 10 11 [START FIGURE 9.7 HERE] 12 13 Figure 9.7: Meridional-depth profiles of zonal-mean potential temperature in the ocean and its rate of change 14 in the upper 2000m of the Global, Pacific, Atlantic, and Indian Oceans. Shown are (a, e, i, m) 15 observed temperature (Argo climatology 2005-2014), (b, f, j, n) bias of the CMIP6 ensemble over this 16 period, and future changes under (c, g, k, o) SSP1-2.6 and (d, h, l, p) SSP5-8.5. No overlay indicates 17 regions with high model agreement, where ≥80% of models agree on the sign of change; diagonal lines 18 indicate regions with low model agreement, where <80% of models agree on the sign of change (see 19 Cross-Chapter Box Atlas.1 for more information). Further details on data sources and processing are 20 available in the chapter data table (Table 9.SM.9). 21 22 [END FIGURE 9.7 HERE] 23 24 25 Different processes drive OHC patterns over a range of timescales. Recent literature has highlighted the role 26 of ocean circulation variability in driving OHC patterns by decomposing the global pattern of OHC change 27 into a combination of added heat due to climate change taken up under fixed ocean circulation (“added 28 heat”), and redistribution of heat associated with changing ocean currents (“redistributed heat”) (Gregory et 29 al., 2016; Bronselaer and Zanna, 2020; Couldrey et al., 2020). Redistributed heat alters regional patterns of 30 heat storage (and carbon storage; Cross-Chapter Box 5.3) (Bronselaer and Zanna, 2020; Couldrey et al., 31 2020; Todd et al., 2020) but does not affect the global OHC. There is medium confidence that decadal 32 variability of the ocean circulation strengthened the rate of ocean warming in the Southern Hemisphere 33 compared to the Northern Hemisphere in the decade from 2005 (Rathore et al., 2020; Wang et al., 2020; Zika 34 et al., 2021). More generally, since 2005, the OHC pattern observed is predominantly due to heat 35 redistribution with regions of both warming and cooling (Zika et al., 2021) (Figure 9.6) but extending 36 analysis back to 1972 shows the importance of added heat setting a large-scale warming pattern with mid- 37 latitude maxima consistent with subduction of water masses, particularly in Southern Hemisphere Mode 38 Waters (Section 9.2.2.3;Figures 9.6, 9.8) (Bronselaer and Zanna, 2020). The longer the analysis window, the 39 more added heat dominates over redistributed heat. This translates into more ocean area with statistically 40 significant warming trends and less area with statistically significant cooling trends (Johnson and Lyman, 41 2020). The region where added heat is most compensated for by redistributed cooling is in the northern 42 North Atlantic basin, where changes in the subpolar gyre circulation and AMOC result in cooling (Section 43 9.2.3.1) (Williams et al., 2015b; Piecuch et al., 2017; Zanna et al., 2019a; Bronselaer and Zanna, 2020). In 44 summary, and strengthening the SROCC assessment, ocean warming is not globally uniform due to patterns 45 of uptake predominantly along known water mass pathways, and due to changing ocean circulation 46 redistributing heat within the ocean (high confidence). 47 48 49 [START FIGURE 9.8 HERE] 50 51 Figure 9.8: Decomposition of ocean simulated ocean heat content and northward heat transport. (a, c, e) Total 52 ocean heat content (0-2000 m) warming rate as observed and simulated by CMIP5 models over the 53 historical period (1951-2011) and under the RCP8.5 future (2011-2060) versus the associated 54 decomposed (b, d, f) added heat contribution (neglecting changes in ocean circulation) to the total 55 (Bronselaer and Zanna, 2020). (g) Relationship between northward heat transport and Atlantic Meridional 56 Overturning Circulation in HighResMIP models (1950-2050) and observations during the RAPID period Do Not Cite, Quote or Distribute 9-23 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 (2004-2018). Further details on data sources and processing are available in the chapter data table (Table 2 9.SM.9). 3 4 [END FIGURE 9.8 HERE] 5 6 7 While heat redistribution reflects changes in ocean circulation and is a useful concept to understand the 8 underlying processes driving OHC patterns, change in ocean heat transport (OHT) arises due to changes in 9 both ocean circulation and ocean temperature and affects regional OHC change. The AR5 did not assess 10 change in OHT and the SROCC (Meredith et al., 2019) only assessed projected OHT increases into the 11 Nordic Seas and the Arctic Ocean. New evidence of increasing northward OHT into the Arctic has been 12 observed in recent decades (Muilwijk et al., 2018; Wang et al., 2019b; Tsubouchi et al., 2021), similar to the 13 SROCC assessment, and consistent with observed increase in OHC in the ice free Arctic ocean (Mayer et al., 14 2019). It is estimated that an increase of 0.021 PW of OHT occurred after 2001 into the Arctic, which is 15 sufficient to account for the recent OHC change in the northern seas (Tsubouchi et al., 2021). However, 16 these trends cannot yet be attributed to anthropogenic forcing due to potential internal variability (Muilwijk 17 et al., 2018; Wang et al., 2019b). New evidence strengthens the case that ENSO and the Northern Annular 18 Mode affect interannual OHT variability (Trenberth et al., 2019) and shows that a slowing AMOC reduces 19 northward OHT in the Atlantic at 26.5N (Bryden et al., 2020) (Section 9.2.3.1, Figure 9.8). Despite a 20 decrease of AMOC northward heat (0.17 PW) and mass (2.5 Sv) transport, OHT has increased toward the 21 Arctic through increased upper northern North Atlantic temperatures and stronger wind-driven gyres 22 (medium confidence) (Section 9.2.3.4, Figure 9.11) (Singh et al., 2017; Oldenburg et al., 2018a). In 23 summary, OHT has increased toward the Arctic in recent decades, which at least partially explains the recent 24 OHC change in the Arctic (medium confidence). 25 26 Major volcanic eruptions have caused interannual to decadal cooling phases within the marked long-term 27 increase in global OHC (Agung in 1963, El Chichón in 1982 and Pinatubo in 1991; Cross-Chapter Box 4.1) 28 (Church et al., 2005; Fasullo et al., 2016; Stevenson et al., 2016; Fasullo and Nerem, 2018). In the first few 29 years following an eruption, heat exchange with the subsurface ocean allows atmospheric cooling to be 30 sequestered into the seasonal thermocline, therefore reducing the magnitude of the peak atmospheric 31 temperature anomaly (Gupta and Marshall, 2018). However, while explosive volcanic eruptions only disturb 32 the Earth’s radiative budget and surface fluxes for a few years, the ocean preserves an anomaly in OHC in 33 the upper 500m (also affecting thermosteric sea level) many years after the eruption (Gupta and Marshall, 34 2018; Bilbao et al., 2019). The anomaly affects the atmosphere through air-sea heat fluxes with surface 35 conditions returning to normal only after several decades (Gupta and Marshall, 2018; Bilbao et al., 2019), or 36 on centennial time-scales in the case of repeated eruptions (Miller et al., 2012a; Atwood et al., 2016; Gupta 37 and Marshall, 2018). In summary, there is medium confidence that oceanic mechanisms buffer the 38 atmospheric response to volcanic eruptions on annual timescales by storing volcanic cooling in the 39 subsurface ocean, affecting ocean heat content and thermosteric sea level on decadal to centennial 40 timescales. 41 42 CMIP5 and CMIP6 models simulate OHC changes that are consistent with the updated observational and 43 improved estimates of OHC over the period 1960 to 2018 (Figures 9.6, 9.7, 9.8), and they replicate the 44 vertical partitioning of OHC change for the industrial era, although with a tendency to underestimate OHC 45 gain shallower than 2000 m and overestimate it deeper than 2000 m (Section 3.5.1.3). The AR5 (Flato et al., 46 2013) assessed that climate models transport heat downward more than the real ocean. Since the AR5, 47 studies have shown that increasing the horizontal resolution of ocean models tends to increase agreement of 48 vertical heat transport with observations as the dependency on ad-hoc choices of eddy parameterizations is 49 relaxed (Griffies et al., 2015; Chassignet et al., 2020). The magnitude of the AMOC and Indonesian 50 Throughflow affect future OHC change, e.g. through overestimated modelled downward heat pumping 51 (Kostov et al., 2014), and there are indications of greater model consistency in these transports at higher 52 resolution (Chassignet et al., 2020; Jackson et al., 2020a) (Figure 9.10). Climate models tend to reproduce 53 the observed added heat, but redistributed heat is less well represented (Figure 9.8) (Bronselaer and Zanna, 54 2020; Dias et al., 2020; Couldrey et al., 2021). Since redistributed heat dominates historical OHC change, 55 historical simulations poorly reproduce regional patterns, but as future OHC change will become dominated Do Not Cite, Quote or Distribute 9-24 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 by added heat, more skill in future modelled OHC patterns is expected (Bronselaer and Zanna, 2020). In 2 summary, climate models have more skill in representing OHC change from added heat than from ocean 3 circulation change (high confidence). Since added heat dominates over redistributed heat on a centennial 4 scale (especially under high emissions scenarios) confidence in future modelled OHC patterns at the end of 5 the 21st century is greater than at decadal scale. 6 7 The SROCC (Bindoff et al., 2019) assessed that the ocean will continue to take up heat in the coming 8 decades for all plausible scenarios, and here this assessment is confirmed with very high confidence. The 9 SROCC reported that compared with the observed changes since the 1970s, the warming of the ocean by 10 2100 would very likely double to quadruple for low emissions scenarios (RCP2.6) and increase 5 to 7 times 11 for high emissions scenarios (RCP8.5). The SROCC also concluded with high confidence that the overall 12 warming of the ocean would continue this century even after radiative forcing and mean surface 13 temperatures stabilize, and SROCC projected that ocean heat content in the 0–2000 m layer will increase 14 from 2017 to 2100 by 0.900±0.345 YJ under RCP2.6 and 2.150±0.540 YJ under RCP8.5. Updating the 15 SROCC estimates with CMIP6 projections gives heat content increases and 17-83% ranges in the 0–2000 m 16 layer from 1995-2014 to 2081-2100 of 1.06 (0.80 – 1.31) YJ, 1.35 (1.08 – 1.67) YJ, 1.62 (1.37 – 1.91) YJ, 17 1.89 (1.60 – 2.29) YJ under scenarios SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, respectively (Figure 9.6, 18 Table 9.1). The two-layer model used here to calculate thermosteric sea level rise (9.SM.4) and tuned for 19 AR6-assessed ECS (Section 7.SM.2), provides consistent 17-83% ranges of 1.18 (0.99 – 1.42) YJ, 1.56 (1.33 20 – 1.86) YJ, 1.90 (1.63 – 2.21) YJ, 2.23 (1.92 – 2.64) YJ under scenarios SSP1-2.6, SSP2-4.5, SSP3-7.0, and 21 SSP5- 8.5, respectively (Table 9.1). Based on both CMIP6 models and the two-layer model, it is likely that, 22 from 1995-2014 to 2081-2100, OHC will increase 2 to 5 times the amount of the 1971-2018 OHC increase 23 under SSP1-2.6, and 4 to 8 times that amount under SSP5-8.5. CMIP6 models show that the dependence of 24 OHC on scenarios begins only after about 2040 (Figure 9.6). 25 26 The patterns of OHC projected by CMIP6 models (Figures 9.6, 9.7) are similar to the CMIP5 projections 27 assessed in the SROCC (Bindoff et al., 2019): faster warming in all water mass subduction regions (e.g., 28 subtropical cells and Mode waters); deeper penetration in the centre of subtropical gyres; slower northern 29 North Atlantic warming due to slowing AMOC; and slower subpolar Southern Ocean warming due upwelled 30 pre-industrial water masses. Decreased aerosol forcing will allow Northern Hemisphere ocean warming to be 31 faster and less dominated by Southern Hemisphere change (Shi et al., 2018; Irving et al., 2019). Since the 32 SROCC, distinguishing between added and redistributed heat has aided in understanding projections 33 (Bronselaer and Zanna, 2020; Couldrey et al., 2020; Dias et al., 2020). The near-term decades will feature 34 patterns strongly influenced by heat redistribution and internal variability (Rathore et al., 2020). 35 Strengthening Southern Hemisphere westerlies are projected except for stringent mitigation scenarios 36 (Bracegirdle et al., 2020) and will cause a northward and downward OHT. There is low agreement in future 37 Southern Ocean warming across model results due to uncertainties in the magnitude of westerly wind 38 changes (Figure 9.4) (Liu et al., 2018; He et al., 2019; Dias et al., 2020; Lyu et al., 2020b) and the degree of 39 eddy compensation of overturning across different parameterisations and resolutions (9.2.3.2) (Beal and 40 Elipot, 2016; Mak et al., 2017; Roberts et al., 2020a). By 2100 however, the OHC change will be dominated 41 by the added heat response, particularly for strong warming scenarios (Garuba and Klinger, 2018; Bronselaer 42 and Zanna, 2020) with added heat following unperturbed water mass pathways in the North Atlantic and 43 Southern Ocean (Couldrey et al., 2020; Dias et al., 2020) (Figure 9.8). There is high confidence that 44 projected weakening of the AMOC (Section 9.2.3.1) will cause a decrease in northward OHT in the northern 45 hemisphere mid-latitudes (Figure 9.8; Sections 9.2.3.1,4.3.2.3; Weijer et al., 2020) associated with a dipole 46 pattern of Atlantic OHC redistributed from northern to low latitudes that may override added heating in the 47 northern North Atlantic (Figures 9.6, 9.7, 9.8). Variations in the degree of AMOC redistributed heat 48 (Menary and Wood, 2018) causes large intermodel spread in SST (Figure 9.3) and OHC (Figure 9.6) change 49 (Kostov et al., 2014; Bronselaer and Zanna, 2020; Couldrey et al., 2020; Todd et al., 2020). In the 700- 50 2000m depth range, CMIP5 and CMIP6 models project the largest warming to be in the North Atlantic Deep 51 Water and Antarctic Intermediate Water (Figure 9.7) while below 2000 m, the North Atlantic cools in many 52 models, and Antarctic Bottom Waters warm (Sallée et al., 2013b; Heuzé et al., 2015). In summary, on 53 decadal timescales, redistribution will dominate regional patterns of OHC change without affecting the 54 globally integrated OHC, but by 2100, particularly under strong warming scenarios, there is high confidence 55 that regional patterns of OHC change will be dominated by added heat entering the sea surface primarily in Do Not Cite, Quote or Distribute 9-25 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 water-mass formation regions in the subtropics, with reduced aerosols increasing the relative rate of 2 Northern Hemisphere heat uptake (medium confidence). 3 4 The SROCC assessed that the warming of the deep ocean is slow to manifest, with multi-century or longer 5 response times, so global OHC (and global mean thermosteric sea level) will continue to rise for centuries 6 (Figures 9.9, 9.30). New studies show that this continuation persists even after cessation of greenhouse gas 7 emissions (Ehlert and Zickfeld, 2018a). Ocean warming will continue even after emissions reach zero 8 because of slow ocean circulation (Larson et al., 2020). Ocean heat content will increase until at least 2300 9 even for low emission scenarios, but with a scenario-dependent rate (Nauels et al., 2017; Palmer et al., 2018) 10 and depends not only on cumulative CO2 emissions, but also on the time profile of emissions (Bouttes et al., 11 2013). Past long-term changes in total OHC illustrate adjustment relevant to expected future changes (Figure 12 9.9). Observational data (Figure 9.9) from ice core rare gas elemental and isotopic ratios document a rise in 13 global OHC relative to the Last Glacial Maximum of >17,000 ZJ (change in mean ocean temperature 14 >3.1oC) (Bereiter et al., 2018; Baggenstos et al., 2019; Shackleton et al., 2019, 2020). This temperature 15 increase is significantly larger than the modelled OHC changes associated with collapse of AMOC alone and 16 tracks rising Southern Ocean SST (Uemura et al., 2018), strengthening of the deep abyssal overturning cell 17 (Du et al., 2020) and increased North Atlantic water in the Southern Ocean (Wilson et al., 2020), 18 underscoring the importance of Antarctic abyssal ventilation on long-term oceanic heat budgets (Section 19 9.2.3.2). An ensemble of four intermediate-complexity models project 10,000-year future responses to CO2 20 emissions (Clark et al., 2016) with SST change peaking around 2300 with varying scenario-dependent 21 magnitude approaching the scale of glacial-to-interglacial changes in paleodata (Figure 9.9). Long-term 22 OHC commitments relative 1850-1900 conditions are 2.6, 9.7, 15.2, 21.6, and 28.0 YJ (with mean ocean 23 temperature change as much as 5.1oC) for emissions of 0, 1280, 2560, and 3840 and 5120 Gt after 2000 CE 24 respectively, with OHC peaking near 4000 CE reflecting whole-ocean warming lagging SST by thousands of 25 years. The exact timing is uncertain subject to rates of high-latitude meltwater input (Van Breedam et al., 26 2020) and circulation time (Gebbie and Huybers, 2019). In summary, there is high confidence that there is a 27 long-term commitment to increased OHC in response to anthropogenic CO2 emissions, which is essentially 28 irreversible on human timescales. 29 30 31 [START FIGURE 9.9 HERE] 32 33 Figure 9.9: Long-term trends of ocean heat content and surface temperature. (a, b) Ice-core rare gas estimates of 34 past mean ocean heat content OHC (ZJ), scaled to global mean ocean temperature (°C) and to steric 35 GMSL (m) (red dashed line) are compared to surface temperatures (black solid line, gold solid line; °C 36 rightmost axis). Southern Ocean SST from multiple proxies in 11 sediment cores and from ice core 37 deuterium excess (Uemura et al., 2018). a) Penultimate glacial interval to last interglacial, 150,000- 38 100,000 yr B2K (Shackleton et al., 2020). b) Last glacial interval to modern interglacial, 40,000-0 yr B2K 39 (Baggenstos et al., 2019; Shackleton et al., 2019). Changes in OHC (dashed lines) track changes in 40 Southern Ocean SST (solid lines). c) Long-term projected (2000 to 12000 CE) changes of OHC (dashed 41 lines) in response to four greenhouse gas emissions scenarios (Clark et al., 2016) scale similarly to large- 42 scale paleo changes but lag projected gobal mean SST (solid lines). d) model simulated 1500-1999 OHC 43 (Gregory et al., 2006) and 1955-2019 observations (Levitus et al., 2012) updated by NOAA NODC. All 44 data expressed as anomalies relative to pre-industrial time. Further details on data sources and processing 45 are available in the chapter data table (Table 9.SM.9). 46 47 [END FIGURE 9.9 HERE] 48 49 50 9.2.2.2 Ocean Salinity 51 52 The AR5 (Rhein et al., 2013) assessed that it was very likely that subsurface salinity changes reflect surface 53 salinity change, and that basin-scale regions of high salinity and evaporation had trended more saline, while 54 regions of low salinity and more precipitation had trended fresher since the 1950s. The SROCC (Bindoff et 55 al., 2019) assessment was consistent with the AR5. Section 2.3.3.2 strengthens evidence that subsurface 56 salinity trends are connected to surface trends (very likely), which are in turn linked to an intensifying Do Not Cite, Quote or Distribute 9-26 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 hydrological cycle (medium confidence) and increasing evidence from updated observational records 2 indicates it is now virtually certain that surface salinity contrasts are increasing. At basin scale, Section 3 2.3.3.2 and the AR5 concur that it is very likely that the Pacific and Southern Ocean have freshened, and the 4 Atlantic has become more saline. Figures 3.25 and 3.27 compare CMIP6 models to salinity observations. 5 6 Globally the mean salinity contrast at near-surface between high- and low- salinity regions increased 0.14 7 [0.07 to 0.20] from 1950 to 2019 (Section 2.3.3.2). At regional scale, the SROCC (Meredith et al., 2019) 8 assessed an Arctic liquid freshwater trend of 600 ± 300 km3 yr–1 (600 ± 200 Gt yr-1) between 1992 and 2012, 9 reflecting changes associated with continental freshwater imports that affect ocean mass (land ice, rivers) as 10 well as changes in sea ice volume. Since the AR5, regional observation-based analyses not assessed in the 11 SROCC further confirm the long-term, large-scale and regional patterns of salinity change, both at the ocean 12 surface and in the subsurface ocean, including almost 120 years of changes in the North Atlantic (Friedman 13 et al., 2017) and 60 years of monitoring in the subpolar North Pacific (Cummins and Ross, 2020b). These 14 longer time-series also provide context to detect large multi-annual change, from 2012 to 2016 in the 15 subpolar North Atlantic, unprecedented over the centennial record (Holliday et al., 2020). In summary, there 16 is high confidence that salinity trends have extended for more than 60-100 years in the regions with long 17 historical observation records such as the North Pacific and the North Atlantic basin. 18 19 While there is low confidence in direct estimates of trends in surface freshwater fluxes (Sections 2.3.1.3.5, 20 8.3.1.1, 9.2.1.2), as discussed in the SROCC (Meredith et al., 2019), observational studies coupled with 21 modelling studies suggest that surface flux changes drive many observed near-surface salinity changes, on 22 top of changes specific to polar regions. Advances in salinity observations (e.g., the Argo program (Riser et 23 al., 2016); SMOS, Aquarius and SMAP (Supply et al., 2018; Vinogradova et al., 2019)), combined with 24 process-studies (SPURS-1/2; (Lindstrom et al., 2015; SPURS-2 Planning Group 2015)) and methodological 25 and numerical advances, have increased understanding of how subsurface salinity anomalies link to surface 26 fluxes, and thus increase confidence that near-surface and subsurface salinity pattern changes since the 1950s 27 are linked to changing surface freshwater fluxes (Zika et al., 2018; Cheng et al., 2020) with an additional 28 contribution from changes in sea-ice and land-ice discharge at high latitudes (Haumann et al., 2016; Purich et 29 al., 2018; Dukhovskoy et al., 2019a; Rye et al., 2020). There is therefore medium confidence in the processes 30 linking surface fluxes to surface and subsurface salinity change. 31 32 Ocean circulation changes also affect salinity, largely on annual to decadal timescales (Du et al., 2019; Liu et 33 al., 2019; Holliday et al., 2020). For instance, in the subpolar North Atlantic, increasing northward transport 34 of “Atlantic waters” entering the subpolar gyre from the South have compensated the salinity decrease 35 expected from increased Greenland meltwater flux since the early 1990s (Dukhovskoy et al., 2016, 2019b; 36 Stendardo et al., 2020). After the mid-2010s the trend reversed towards a broad freshening, the largest in 120 37 years, in the North Atlantic (Holliday et al., 2020). The long term freshening in the Pacific Ocean has also 38 been subject to decadal variability, such as a marked salinification since 2005 associated with increased 39 surface fluxes (Li et al., 2019a). Local salinity anomalies forced by water cycle intensification can be 40 weakened by rapid exchange between basins with opposing trends, such as by water-mass exchange in 41 shallow wind-driven cells between the tropics and the subtropics (Levang and Schmitt, 2020). Similarly, 42 eddy exchanges between neighbouring gyres can partly counterbalance decadal time scale long-term 43 subpolar freshening and affect deep convection (Levang and Schmitt, 2020). There is high confidence that, at 44 annual to decadal timescales, regional salinity changes are driven by ocean circulation change superimposed 45 on longer term trends. 46 47 CMIP5 historical simulations have patterns similar to, but greater spatial variability than, observed estimates 48 and correspondingly smaller amplitudes in the multi-model mean (Durack, 2015; Cheng et al., 2020; Silvy et 49 al., 2020a). Section 3.5.2.1 reports, however, that the fidelity of ocean salinity simulation has improved in 50 CMIP6, and near-surface and subsurface biases have been reduced (medium confidence), though the 51 structure of the biases strongly reflects those of CMIP5. At regional scale, salinity biases are at least partially 52 a result of inaccurate ocean dynamics (Levang and Schmitt, 2020). Despite the regional limitations, Section 53 3.5.2.2 assesses that at the global scale it is extremely likely that human influence has contributed to observed 54 surface and subsurface salinity changes since the mid-20th century (strengthened from the very likely AR5 55 assessment). Do Not Cite, Quote or Distribute 9-27 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 2 The SROCC (Bindoff et al., 2019) assessed that projected salinity changes in the subsurface ocean reflect 3 changes in the rates of formation of water masses or their newly formed properties. Additional consistent 4 newer evidence based on CMIP5 and regional climate models confirms that 21st century projections adhere 5 to the fresh gets fresher, salty gets saltier paradigm, through subduction of freshening high latitude waters 6 into the ventilated water-masses in both hemispheres in the Pacific, Indian and Southern Ocean, especially 7 the Arctic and upper Southern Ocean, and saltier subtropical and Mediterranean surface waters lead to saltier 8 pycnoclines and North Atlantic mode water (Metzner et al., 2020; Parras-Berrocal et al., 2020; Silvy et al., 9 2020a; Soto-Navarro et al., 2020). Overall, projections confirm the SROCC assessment that fresh ocean 10 regions will continue to get fresher and salty ocean regions will continue to get saltier in the 21st century 11 (medium confidence). 12 13 14 9.2.2.3 Water Masses 15 16 Water masses refer to connected bodies of ocean water, formed at the ocean surface with identifiable 17 properties (temperature, salinity, density, chemical tracers) resulting from the unique formation conditions of 18 the overlying atmosphere and/or ice, before being transferred (subducted) to the deeper ocean below the 19 surface turbulent layer. As water masses subduct, they ventilate the subsurface ocean, transferring 20 characteristics acquired at the ocean surface to the subsurface. By integrating surface flux changes, water 21 masses provide higher signal-to-noise ratios for detecting and monitoring climate change than surface fluxes 22 (Bindoff and McDougall, 2000; Durack and Wijffels, 2010; Silvy et al., 2020b). 23 24 SubTropical Mode Waters (STMW) ventilate the main thermocline of the ocean at mid- to low-latitudes and 25 have circulation timescales away from the surface of the order of years to decades. The SROCC (Bindoff et 26 al., 2019) reported that warming in the subtropical gyres penetrates deeper than in other gyres, following the 27 density surfaces in these gyres. Consistently, we assess that STMW have deepened worldwide, with greatest 28 deepening in the Southern Hemisphere (high confidence) (Häkkinen et al., 2016; Desbruyères et al., 2017). 29 Subsurface warming in the Northern Hemisphere STMW is larger than at the surface (Sugimoto et al., 2017) 30 because they are formed in wintertime western boundary current extensions, where surface warming is larger 31 than the global average (Section 9.2.1.1). Variability in STMW thickness or temperature has a large imprint 32 on ocean heat content (Section 9.2.2.1) (Kolodziejczyk et al., 2019). STMW are observed to be freshening in 33 the North Pacific and to be associated with increased salinity in the North Atlantic (Oka et al., 2017; Silvy et 34 al., 2020a), with large decadal variability (Oka et al., 2019; Wu et al., 2020). Anthropogenic temperature and 35 salinity changes in the STMW layer are projected to intensify in the future, with emergence from natural 36 variability around 2020 to 2040 (Silvy et al., 2020a). 37 38 SubAntarctic Mode Water (SAMW) and Antarctic Intermediate Waters (AAIW) form at the Southern Ocean 39 surface directly north of the Antarctic Circumpolar Current and ventilate the upper 1000 m of the Southern 40 Hemisphere subtropics. The SROCC (Meredith et al., 2019) reported a freshening of these water masses 41 between 1950 and 2018, and they are projected to have the largest subsurface temperature increase of the 42 Southern Hemisphere oceans, along with a continued freshening, in the 21st century. The SROCC connected 43 SAMW and AAIW to Southern Ocean temperature changes as the large Southern Ocean surface heat uptake 44 is circulated and mixed along with these water masses (high confidence). Close to its formation region, 45 SAMW is predominantly affected by air-sea flux changes, while further northward it is influenced by wind- 46 forced changes (Meredith et al., 2019). New evidence shows that a change in SAMW heat content over the 47 last decade is primarily attributable to its thickening (Kolodziejczyk et al., 2019). Over the past decade, the 48 SAMW and AAIW volumes have changed by thickening of the lighter and thinning of the denser parts of 49 SAMW and AAIW, leading to lightening of these ventilated ocean layers overall (Hong et al., 2020; Portela 50 et al., 2020). Over the last decade, there is limited evidence of increased subduction of SAMW due to 51 deepening mixed layers in the SAMW formation region (9.2.1.3) (Qu et al., 2020). Climate models from 52 CMIP3 to CMIP5 generally simulated shallower and lighter SAMW and AAIW than is observed (Flato et 53 al., 2013). New analysis of CMIP5 models suggests that the freshening of these water masses is one of the 54 most prominent projected salinity changes in the world ocean, and that this freshening emerged from internal 55 variability as early as the 1980–1990s (Silvy et al., 2020b). Do Not Cite, Quote or Distribute 9-28 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Trends in North Atlantic Deep Water (NADW) are obscured by decadal variability (Rhein et al., 2013; 2 Bindoff et al., 2019). The AR5 (Rhein et al., 2013) assessed that it is very likely that the temperature, salinity, 3 and formation rate of the Upper NADW (formed by deep convection in the Labrador and Irminger Seas) is 4 dominated by strong decadal variability related to the North Atlantic Oscillation (NAO) and it is likely that 5 Lower NADW (formed in the Nordic Seas and supplied to the North Atlantic by deep overflows over the 6 sills between Scotland and Greenland) cooled from 1955 to 2005. New insights from observations have 7 emphasized the stability of the deep overflows associated with Lower NADW (Hansen et al., 2016; 8 Jochumsen et al., 2017; Østerhus et al., 2019) and even slight warming in the Faroe Bank Channel (Hansen 9 et al., 2016). As a result, the AR5 assessment that Lower NADW likely cooled between 1955 and 2005 is 10 revised to: it is likely that any observed changes in temperature, salinity, and formation rate of the Lower 11 NADW are dominated by decadal variability. For CMIP5 models it was shown that AMOC variability is 12 linked to variability in NADW formation (Heuzé, 2017) and projected AMOC decline to decreased NADW 13 formation (both Lower NADW and Upper NADW) (Heuzé et al., 2015). For CMIP6 models, projected 14 AMOC decline is also associated with a decline in NADW formation (Reintges et al., 2017; Weijer et al., 15 2020). The link between AMOC and NADW formation appears insensitive to the large range in model bias 16 in NADW water mass characteristics (Heuzé, 2017). Many models may overestimate deep water formation 17 in the Labrador Sea, but at least one new model is consistent with recent OSNAP observations showing very 18 weak overturning in the western subpolar gyre, where Labrador Sea Water is formed (Menary et al., 2020a). 19 CMIP6 models show a reduced bias in NADW properties compared to CMIP5 models, but still feature 20 varying locations of deep convection in the subpolar gyre: some convect only in the Labrador Sea (6/35 21 models), most in both the Labrador and Irminger Seas (26/35 models; as is observed), and some only in the 22 Irminger Sea (3/35 models), but in general the area where deep convection takes place has expanded relative 23 to CMIP5, which appears unrealistic (Heuzé, 2021). Models with most deep convection in the subpolar gyre 24 feature the smallest bias in NADW characteristics, partly associated with NADW formed in the Nordic Seas 25 (as observed) being largely unable to leave the area (Heuzé, 2021) due to inaccurate overflows (Danabasoglu 26 et al., 2010; Deshayes et al., 2014; Wang et al., 2015b). Despite the wide range in model bias, it remains very 27 likely that any long-term (multi-decadal or longer) decrease in AMOC is accompanied by a decline in 28 NADW formation, associated with lighter densities in the northern North Atlantic and Arctic basins. 29 30 The SROCC (Meredith et al., 2019) assessed that the global volume of Antarctic Bottom Water (AABW) 31 had decreased and warmed since the 1980s, most noticeably near Antarctica. The SROCC also noted 32 freshening in the Indian and Pacific sectors of the Southern Ocean and a higher rate of freshening in the 33 Indian Sector from the 2000s to 2010s than from the 1990s to 2000s (low confidence). Since the SROCC, 34 freshening of Indian Ocean AABW from 1974 to 2016 has been revealed (Aoki et al., 2020). Additionally, 35 interannual to decadal variability in AABW has been quantified to be larger than previously thought in terms 36 of temperature, salinity and thickness, and in volume transport (Abrahamsen et al., 2019; Purkey et al., 37 2019b; Gordon et al., 2020; Silvano et al., 2020). Multidecadal-to-centennial modes of variability could have 38 driven the observed trends of the lower cell over the past decades via the opening of a Weddell Sea polynya 39 (Zhang et al., 2019b), although other studies find it contributed minimally to the observed abyssal warming 40 (Zanowski et al., 2015; Zanowski and Hallberg, 2017). Therefore, there is limited evidence and low 41 agreement in the role of open ocean polynyas in driving past decadal observed trends of AABW. Beyond 42 variability, all observational, theoretical, and numerical evidence supports the SROCC assessment that 43 formation and export of AABW will continue to decrease due to warming and freshening of surface source 44 waters near the Antarctic continent. Consistent with Section 9.2.3.2, confidence in this assessment is 45 increased to medium confidence compared to the SROCC. 46 47 Circumpolar Deep Water (CDW) lies in the Southern Ocean and forms by mixing of North Atlantic Deep 48 Water and Antarctic Bottom Water (Talley, 2013). The SROCC (Meredith et al., 2019) assessed with low 49 confidence that mean southward and upward CDW transport is linked to decadal wind variability (Section 50 9.2.3.2), and that CDW has warmed south of the Antarctic Circumpolar Current (ACC) in the past decades. 51 New evidence reinforces the SROCC assessment: changes in Southern Ocean wind stress have been 52 confirmed to drive variability and increase the large-scale southward CDW transport (Waugh et al., 2019). In 53 addition, growing evidence suggests that the upper ocean stratification increase in the subpolar Southern 54 Ocean since the 1970s (Section 9.2.1.3) has reduced the volume of CDW that is mixed to the surface, 55 causing subsurface CDW warming (Bronselaer et al., 2020; Haumann et al., 2020; Jeong et al., 2020; Do Not Cite, Quote or Distribute 9-29 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Moorman et al., 2020). Large regions of the Antarctic shelves are currently isolated from warm CDW 2 (Thompson et al., 2018; Jourdain et al., 2020). The SROCC (Meredith et al., 2019) assessed that subsurface 3 warming extends close to Antarctica and has co-occurred with shoaling of the CDW since the 1980s, 4 influencing the continental shelf most in the Amundsen-Bellingshausen Seas, Wilkes Land, and the Antarctic 5 Peninsula. New evidence since the SROCC reinforces confidence in the importance of the role of winds in 6 transporting heat associated with CDW to continental shelves and ice cavities in the Amundsen- 7 Bellingshausen Seas (Dotto et al., 2019) and via variable small-scale undercurrents to the Shirase Glacier 8 Tongue in East Antarctica (Hirano et al., 2020; Kusahara et al., 2021). There is limited evidence that 9 increased greenhouse gas forcing has caused a slight mean change of the local winds from 1920-2018 10 facilitating CDW heat intrusion onto the Amundsen-Bellingshausen continental shelf and ice-shelf melt 11 (Holland et al., 2019). Multiple lines of observational, numerical, theoretical, and paleo evidence provide 12 high confidence that changes in wind pattern (Spence et al., 2014; Dotto et al., 2019; Holland et al., 2019), 13 increased ice-shelf melt (Golledge et al., 2019a; Moorman et al., 2020), reduction in sea-ice 14 production (Timmermann and Hellmer, 2013; Obase et al., 2017), and eddies (Stewart and Thompson, 2015; 15 Thompson et al., 2018) can facilitate access of CDW to the sub-ice-shelf cavities (9.4.2.1). However, there is 16 low confidence in the quantitification, importance and the ability of present models, especially at coarse 17 resolution, to project changes in each of these processes (9.4.2.2). Some studies have projected a possible 18 shift from cold to warm sub-ice shelf cavities causing a sudden flush of warm water underneath ice shelves, 19 but there is low confidence both in the driving processes and the threshold to trigger the shift (Box 9.4) 20 (Hellmer et al., 2012, 2017; Silvano et al., 2018; Hazel and Stewart, 2020). 21 22 23 9.2.3 Regional Ocean Circulation 24 25 9.2.3.1 Atlantic Meridional Overturning Circulation 26 27 The Atlantic meridional overturning circulation (AMOC) is the main overturning current system in the South 28 and North Atlantic Oceans. It transports warm upper-ocean water northwards, and cold, deep water 29 southwards, as part of the global ocean circulation system (Section 2.3.3.4.1). AMOC changes influence 30 global ocean heat content and transport (Section 9.2.2.1); global ocean anthropogenic carbon uptake changes 31 and climate sensitivity (Cross-Chapter Box 5.3); and dynamical sea level (Section 9.2.4). Since the 32 AR5/SROCC, confidence in modelled and reconstructed AMOC has decreased due to new observations and 33 model disagreement. Confidence in modelled AMOC evolution during the 20th century, the magnitude of 34 21st century AMOC decline, and the possibility of an abrupt collapse before 2100 have been revisited. 35 36 The AR5 (Flato et al., 2013) found that the mean AMOC strength in CMIP5 models ranges from 15 to 30 Sv 37 for the historical period. The multi-model mean overturning at 26N in CMIP5 and CMIP6 is comparable to 38 the RAPID measurements (Reintges et al., 2017), but the inter-model spread in CMIP6 is as large (10-31 Sv) 39 as in CMIP5 ((Weijer et al., 2020); Section 3.5.4). Biases in simulations of the present day AMOC and 40 associated deep convection in the subpolar gyre and Nordic Seas were large in CMIP5 models with many 41 models exhibiting ocean convection that is too deep, over too large an area, too far south and occurring too 42 frequently (Heuzé, 2017) (Section 9.2.1.3; Figure 9.5) related to biases in sea-ice extent, overflows, and 43 freshwater forcing (Deshayes et al., 2014; Wang, Legg and Hallberg, 2015). As a result, the AMOC in 44 CMIP5 was nearly always too shallow, with too weak a temperature contrast between the northward and 45 southward flowing branches. Deep convection errors are still large in CMIP6 and the shallow bias in AMOC 46 persists (Weijer et al., 2020; Heuzé, 2021). Since the AR5, there is emerging evidence that enhancing 47 horizontal resolution can reduce longstanding climate model biases in AMOC strength, where the magnitude 48 and profile of northward heat transport at 26N become more comparable to observations (Chassignet et al., 49 2020; Roberts et al., 2020a). The sensitivity of the AMOC to ocean resolution, however, is model-dependent 50 and can be positive as well as negative (Roberts et al., 2020b). An increase in AMOC strength at 26N with 51 higher resolution in the ocean component has been associated with too strong (deep) convection in the 52 subpolar gyre and too deep winter mixed layers (Jackson et al., 2020a), which occurs in most CMIP6 models 53 that are unable to overflow deep water formed in the Nordic Seas across the Greenland-Iceland-Scotland 54 Ridge. Thus models with a correct AMOC strength may do so by compensating a lack of deep water outflow 55 from the Nordic Seas through too much deep convection and deep-water formation in the Labrador and Do Not Cite, Quote or Distribute 9-30 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Irminger Seas (Heuzé, 2021). 2 3 Models and paleo-reconstructions have often assumed a close relation between the AMOC and deep 4 convection in the Labrador Sea, and Labrador Sea convection variability has been interpreted as connecting 5 to AMOC variability. Observational studies have been inconclusive on whether this relation exists (Buckley 6 and Marshall, 2016). New insight from observed overturning in the eastern and western subpolar gyre in the 7 North Atlantic in OSNAP (Lozier et al., 2019; Petit et al., 2020) reveal that 15.6±3.1Sv takes place north of 8 the OSNAP array between Greenland and Scotland, with only 2.1±0.9 Sv of overturning occurring across the 9 Labrador Sea as found with the OSNAP 53oN array spanning the mouth, explicitly questioning the validity 10 of the Labrador Sea convection-AMOC link (Lozier et al., 2019). Although their results are derived from 11 only the first 21 months of data of monitoring since 2014, hydrographic observations during 1990-1997 12 previously found small overturning (1-2 Sv) in the Labrador Sea (Pickart and Spall, 2007). On the other 13 hand, previous estimates of Labrador Sea Water formation (obtained with different techniques) suggest 14 larger overturning (Haine et al., 2008). Part of this controversy could be explained if a large fraction of 15 newly formed Labrador Sea Water is not exported from the Labrador Sea. The OSNAP observations are 16 supported by previous hydrographic measurements in showing strong east-west symmetry in isopycnal slope 17 in the Labrador Sea in periods of both strong and weak convection, implying compensating northward and 18 southward transport above and below the potential density surface that separates the upper and lower 19 overturning limbs (Lozier et al., 2019), despite large deep convection variability (Yashayaev, 2007; 20 Yashayaev and Loder, 2016). New observations of deep winter mixing in the Irminger Basin (de Jong et al., 21 2018; Josey et al., 2019) support the assertion that the Irminger Sea, in addition to the Nordic Seas (Chafik 22 and Rossby, 2019), are the main sources of overturning in the eastern subpolar gyre, consistent with OSNAP 23 (Petit et al., 2020). It is unclear to what extent models are in disagreement with this view of overturning in 24 the subpolar gyre, as a direct comparison with OSNAP in terms of partitioning the overturning in a western 25 and eastern part is lacking for most models, with a notable exception (Menary et al., 2020a). The analysis of 26 water mass formation in CMIP6 models (Heuzé, 2021); the analysis between Labrador Sea water formation 27 and AMOC in a suite of ocean-only models (Danabasoglu et al., 2014); the fact that when the OSNAP 28 observing system design was tested in an eddy-permitting ocean model, almost equal amounts of overturning 29 in the western and eastern subpolar gyre were found (Lozier et al., 2017), give rise to considerable 30 uncertainty over the models’ veracity in simulating the overturning partitioning between east and west and 31 the role of various drivers of AMOC variability. Disagreement between models and OSNAP observations 32 may decrease in higher-resolution models (Menary et al., 2020a). In summary, multiple lines of evidence 33 provide medium agreement between models and observations on drivers of change and variability in the 34 AMOC and in particular the role of Labrador Sea deep convection in constituting AMOC variability. 35 36 The AMOC is a potential driver of Atlantic Multidecadal Variability (AMV), but there is new evidence that 37 anthropogenic aerosol changes have contributed to observed AMV changes, and that underestimation of the 38 magnitude and duration of AMV changesin CMIP5 is tempered in CMIP6 (Section 3.7.7, Annex IV.2.7). 39 Comparison of observed AMOC variability at the RAPID section with modelled variability reveals that 40 CMIP5 models appear to largely underestimate the interannual and decadal timescale variability (Roberts et 41 al., 2014; Yan et al., 2018), and similar results are found when analysing CMIP6 models (Section 3.5.4.1). 42 By underestimating the multi-decadal AMOC-AMV link and other low-frequency AMOC variability climate 43 models also underestimate internal variability in subpolar SSTs that feedback on the North Atlantic 44 Oscillation (NAO), causing the NAO to lack variability on multidecadal timescales (Kim et al., 2018). 45 Despite the role of the AMOC in generating AMV through subsurface temperatures in antiphase with SST 46 and downward heat fluxes into the ocean that anticorrelate with SSTs (Zhang et al., 2019d), it is generally 47 accepted that AMOC forcing of SST variability exists alongside stochastic wind forcing and external forcing 48 by aerosols (Bellomo et al., 2018; Haustein et al., 2019; O’Reilly et al., 2019; Wills et al., 2019). 49 50 The SROCC (Collins et al., 2019) assessed that in situ observations (2004–2017) and sea surface 51 temperature reconstructions indicate that the AMOC has weakened relative to 1850–1900 (medium 52 confidence). However, the SROCC also assessed that there is insufficient data to quantify the magnitude of 53 the weakening, or to properly attribute it to anthropogenic forcing, due to the limited length of the 54 observational record. Here, this assessment is adjusted to low confidencein the weakening as also discussed 55 in Sections 2.3.3.4.1 and 3.5.4.1. The CMIP5 multi-model mean showed no 20th century trend in the AMOC Do Not Cite, Quote or Distribute 9-31 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 (Cheng et al., 2013). The CMIP6 multi-model mean even slightly opposes the reconstructed decline due to a 2 strong increase in the 1940-1985 period (Menary et al., 2020c; Weijer et al., 2020), thought to be in response 3 to aerosol forcing (Section 3.5.4.1), followed by a smaller decline since the nineties. Also, agreement 4 between different proxy-based reconstructions is weak in many details (Moffa‐Sánchez et al., 2019) and 5 questions can be raised regarding various proxies used in reconstructions (Section 2.3.3.4.1). For instance, 6 SST-based proxies can be influenced by atmospheric and other processes acting on different timescales 7 (Moffa‐Sánchez et al., 2019; Jackson and Wood, 2020). In addition, many proxies are indirect and based on 8 AMOC-related processes assumed to be similar as found in models, such as the link between AMOC and 9 Labrador Sea convection, which has been questioned recently (see above). In addition, the subpolar gyre 10 from which many AMOC-proxies are taken may vary independently of the AMOC with rather similar 11 patterns in SST and ocean heat content driven by wind variability (Williams et al., 2014; Piecuch et al., 12 2017). Finally, a new dynamic reconstruction of the Atlantic inflow to the Nordic Seas suggest no slowdown 13 over the past 70-100 years (Rossby et al., 2020), in contrast to a new compilation of proxy reconstructions 14 which suggests that the AMOC is presently in its weakest state in the last millennium (Caesar et al., 2021), 15 reinforcing the evidence that motivated the previous SROCC assessment. Section 3.5.4.1 also questions the 16 veracity of the models’ forced AMOC response during the twentieth century. Given the large discrepancy 17 between modelled and reconstructed AMOC in the twentieth century and the uncertainty over the realism of 18 the 20th century modelled AMOC response (Section 3.5.4.1), we have low confidence in both. 19 20 The strength of the AMOC has been measured directly since 2004 using the RAPID Array (Smeed et al., 21 2018) (Section 2.3.3.4.1). RAPID-based estimates show a large amount of variability compared to CMIP 22 models (Roberts et al., 2014). Observed changes since 2004 are too short for the evaluation of a long-term 23 trend given the decadal scale internal variability (Section 2.3.3.4.1). Nevertheless, Smeed et al. (2018) argue 24 that between 2007 and 2011 the AMOC shifted to a state of reduced overturning; decreasing from 18.8 Sv 25 between 2004 and 2008 to 16.1 Sv after 2008. A shift in AMOC strength of this magnitude is not captured 26 by CMIP5 and CMIP6 models, which generally underestimate interannual to decadal AMOC variability 27 (Section 3.5.4.1). Additional evidence since SROCC also raises the inconsistency between the RAPID 28 weakening in the 3000-5000 m depth range and the relative constancy of deep overflows from the Arctic 29 (Østerhus et al., 2019), implying that the recent decrease in AMOC at 26.5ºN (Smeed et al., 2018) is not 30 caused by overflow weakening or reduced overturning in the Nordic Seas, although the weakening occurred 31 almost exclusively in the 3000 – 5000 m depth range associated with a reduction of Lower NADW (Section 32 9.2.2.3). It is unclear what causes a weakening of the deepest limb of the AMOC at 26.5ºN, if the main 33 sources for this flow farther north remain constant. Various estimates of AMOC and associated heat 34 transport suggest an increase since the 1940s with a subsequent decrease since the 1990s (Section 2.3.3.4.1), 35 supported by ocean reanalysis (Jackson et al., 2019), forced ocean model simulations (Robson et al., 2012; 36 Danabasoglu et al., 2016) and CMIP6 simulations (Menary et al., 2020b). This suggests that the observed 37 AMOC-shift between 2007 and 2011 may be part of a longer-term decrease (medium confidence), which has 38 been attributed to be part of multiannual variability (Rhein et al., 2019). 39 40 41 [START FIGURE 9.10 HERE] 42 43 Figure 9.10: AMOC strength in simulations and sensitivity to resolution and forcing. (Top left) AMOC magnitude 44 in PMIP experiments. (Top right) Time series of AMOC from CMIP5 and CMIP6 based on (Menary et 45 al., 2020c). (Bottom left) Percent change in AMOC strength per year at different resolutions over the 46 1950-2050 period with colours for model families (Roberts et al., 2020b). (Bottom right) A compilation 47 (Jackson and Wood, 2018) of percentage changes in the simulated AMOC after applying an additional 48 freshwater flux in the subpolar North Atlantic at the surface for a limited time (de Vries and Weber, 2005; 49 Stouffer et al., 2006; Yin and Stouffer, 2007; Jackson, 2013; Liu and Liu, 2013; Jackson and Wood, 50 2018; Haskins et al., 2019). Symbols indicate whether the AMOC recovers within 200 years (circles), is 51 starting to recover (upwards arrow) or does not recover within 200 years (downwards arrow). Symbol 52 size indicates rate of freshwater input. Further details on data sources and processing are available in the 53 chapter data table (Table 9.SM.9). 54 55 [END FIGURE 9.10 HERE] 56 Do Not Cite, Quote or Distribute 9-32 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 The SROCC (Collins et al., 2019) found that the AMOC will very likely weaken over the 21st century. In 2 CMIP6 projections, the modelled decline starting in the 1990s continues in all future projections, almost 3 independent of the forcing scenario until about 2060, after which low emission scenarios show stabilization, 4 while high-emission scenarios continue to exhibit AMOC decline (Figure 9.10) (Menary et al., 2020b, 5 Weijer et al., 2020).Despite differences in overall AMOC strength, location and latitude of deep convection, 6 sea-ice and SST bias and representation of deep overflows, the model projections are qualitatively similar. 7 This agreement suggests that AMOC decline may be governed by large-scale constraints independent of the 8 details of the models. In theoretical models of the thermohaline circulation, the circulation strength is 9 proportional to a density or pressure difference between the subpolar North Atlantic and subtropical South 10 Atlantic (Kuhlbrodt et al., 2007; Weijer et al., 2019). In all models, the north-south pressure gradient 11 decreases in the 21st century, as subpolar waters warm faster than subtropical waters and an enhanced 12 hydrological cycle drives freshening at subpolar latitudes, while subtropical latitudes feature more 13 evaporation and salinification (Section 9.2.1). As a result, surface waters at subpolar latitudes become more 14 buoyant and more stable, so that deep water formation driving the AMOC declines (Section 9.2.1.3). 15 Projected AMOC decline by 2100 ranges from 24% (4-46%) in SSP1-2.6 to 39% (17-55%) in SSP5-8.5 16 (medium confidence; Section 4.3.2.3). Note that these ranges are based on ensemble means of individual 17 models, largely smoothing out internal variability. If single realisations are considered the ranges become 18 larger, lowering especially the low end of the range (Section 4.3.2.3). In summary, it is very likely that 19 AMOC will decline in the 21st century, but there is low confidence in the models’ projected timing and 20 magnitude. In addition, freshwater from the melting of the Greenland ice sheet (Sections 9.4.1.3, 9.4.1.4) 21 could further enhance the future weakening of the AMOC in the 21st century (Collins et al., 2019; Golledge 22 et al., 2019). 23 24 Both the AR5 (Collins et al., 2013) and the SROCC (Collins et al., 2019) assessed that an abrupt collapse of 25 the AMOC before 2100 was very unlikely, but the SROCC added that by 2300 an AMOC collapse was as 26 likely as not for high-emission scenarios. The SROCC also assessed that model-bias may considerably affect 27 the sensitivity of the modelled AMOC to freshwater forcing. Tuning towards stability and model biases 28 (Valdes, 2011; Liu et al., 2017; Mecking et al., 2017; Weijer et al., 2019) provides CMIP models a tendency 29 toward unrealistic stability (medium confidence). By correcting for existing salinity biases, Liu et al. (2017) 30 demonstrated that AMOC behaviour may change dramatically on centennial to millennial timescales and that 31 the probability of a collapsed state increases. None of the CMIP6 models features an abrupt AMOC collapse 32 in the 21st century, but they neglect meltwater release from the Greenland ice sheet and a recent process 33 study reveals that a collapse of the AMOC can be induced even by small-amplitude changes in freshwater 34 forcing (Lohmann and Ditlevsen, 2021). As a result, we change the assessment of an abrupt collapse before 35 2100 to medium confidence that it will not occur. 36 37 38 9.2.3.2 Southern Ocean 39 40 The changing Southern Ocean circulation system exerts a strong influence on the global climate by 41 modulating (i) global ocean heat content (Section 9.2.2.1); (ii) global ocean anthropogenic carbon uptake 42 (Cross-chapter Box 5.3); global ocean overturning circulation (Section 9.2.3.1); climate sensitivity (Section 43 7.4.4 and Cross-chapter Box 5.3); (iii) sea level through basal melt of ice shelves (9.4.2); and Southern 44 Hemisphere sea-ice cover (Section 9.3.2). 45 46 The SROCC (Meredith et al., 2019) had low confidence in all CMIP5-based model projections due to their 47 inability to explicitly resolve eddy processes and their inability to properly consider future meltwater change 48 from the Antarctic Ice Sheet. These limitations of climate models to represent the Southern Ocean persist 49 due to most CMIP6 models still using parameterized mesoscale eddy processes that are limited in projecting 50 the future response of the horizontal and vertical circulation under climate warming, and due to the 51 continued absence of active ice shelf and ice sheet coupling in the CMIP6 model suite, therefore ignoring 52 basal meltwater and calving feedback on the circulation (Meredith et al., 2019). In addition, two important 53 limitations of CMIP6 models of the Southern Ocean involve processes that were not assessed in the SROCC. 54 First, the poor representation of dense overflows causes most of the Antarctic Bottom Water (AABW) to be 55 formed by spurious open ocean convection rather than by dense overflows from the Antarctic continental Do Not Cite, Quote or Distribute 9-33 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 shelves that feed the lower overturning cell (Snow et al., 2015; Dufour et al., 2017; Heuzé, 2021). Second, 2 Antarctic continental shelf waters are poorly simulated because potentially important controlling 3 mechanisms tend to be too small and transient to observe and resolve in CMIP ocean models. These small 4 processes include the heterogeneity of observed sub-ice shelf melt with warm water driving narrow basal 5 channels that cut underneath the ice (Drews, 2015; Alley et al., 2016; Marsh et al., 2016; Milillo et al., 6 2019); eddies and tides (Stewart et al., 2018; Jourdain et al., 2019; Hausmann et al., 2020), which can drive 7 Circumpolar Deep Water (CDW) onto the continental shelves or dynamically increase melting (Section 8 9.2.3.6); and feedback mechanisms between ocean, atmosphere and cryosphere that can weaken or amplify 9 initial perturbations (Donat-Magnin et al., 2017; Spence et al., 2017; Turner et al., 2017; Silvano et al., 2018; 10 Webber et al., 2019; Hazel and Stewart, 2020). In addition, the Southern Ocean in CMIP5 and CMIP6 11 models exhibit surface temperature biases (Section 9.2.1.1), which have been linked in CMIP5 model to 12 errors in atmospheric model cloud-related short-wave radiation (Hyder et al., 2018) and are somewhat 13 improved in HighResMIP models (Figure 9.3). In summary, there is high confidence that future change in 14 the subpolar Southern Ocean region including sea-ice cover and ocean temperature change on Antarctic 15 continental shelves depends on feedback mechanisms involving the ocean, atmosphere and cryosphere that 16 are poorly understood and not represented in the current generation of climate models. This results in large 17 uncertainty and low confidence in the future sea-ice cover (Section 9.3.2) and in ocean temperature change 18 on the Antarctic continental shelf (Section 9.4.2.3). 19 20 Despite these challenges, the CMIP6 ensemble does represent the main Southern Ocean circulation 21 characteristics; the simulated Antarctic Circumpolar Current (ACC) transport is generally lower than 22 observation-based values but consistent when considering ensemble spread and the inter-model spread in 23 ACC transport has greatly reduced from previous generations of climate models from CMIP3 to CMIP6 24 (Beadling et al., 2019, 2020). The structure (but not the magnitude) of the two-cell zonally-averaged 25 overturning is captured by most CMIP6 models (Russell et al., 2018; Beadling et al., 2019). In addition, 26 while issues remain, CMIP6 climate models show clear improvements in their representation of AABW 27 compared to CMIP5: several models correctly represent or parameterise Antarctic shelf processes, fewer 28 models exhibit Southern Ocean deep convection, bottom density biases are reduced, and abyssal overturning 29 is more realistic (Heuzé, 2021). In terms of atmospheric wind forcing, CMIP6 models show an improvement 30 compared to CMIP5 models with an overall reduction in the equatorward bias of the annual mean westerly 31 jet from 1.9° in CMIP5 to 0.4° in CMIP6, but in contrast they show no such overall improvements for their 32 representation of the Amundsen Sea Low (Bracegirdle et al., 2020; Lyu et al., 2020a), which can be critical 33 in driving variability of water-masses on the Antarctic continental shelf in west Antarctica, the Weddell Sea 34 or the Ross Sea(Holland et al., 2019; Silvano et al., 2020). 35 36 The SROCC (Meredith et al., 2019) established that while trends in the atmospheric forcing of the Southern 37 Ocean have been dominated by a strengthening of the southern hemisphere westerly winds in recent decades, 38 there is medium confidence that ACC transport is weakly sensitive to changes in winds. It also reported that 39 instead of increasing the mean ACC transport, additional energy input associated with increased wind stress 40 cascades into the eddy field (medium confidence). In contrast with the AR5 assessment (Rhein et al., 2013), 41 the SROCC evaluated that it was unlikely that there has been a net southward migration of the mean ACC 42 position over the past 20 years. There is no additional evidence to revisit the SROCC assessment on wind 43 sensitivity. However, new evidence does suggest that air-sea buoyancy forcing associated with idealised 44 4xCO2 forcing leads to an increase in ACC transport (Shi et al., 2020) (limited evidence). The SROCC noted 45 that if the general strengthening in westerly winds is sustained, then it is very likely that the eddy field will 46 continue to increase in intensity, and that is likely that the mean position and strength of the ACC will remain 47 only weakly sensitive to winds. In the future, the strength of the Southern Hemisphere westerly wind jet 48 results from a competition between decrease due to ozone hole recovery and increase due to increased 49 radiative forcing (Section 4.3.3.1). This competition results in an increased atmospheric jet by 2100 50 compared to present day under SSP2-4.5, SSP3-7.0, and SSP5-8.5, but a decreased jet by 2100 under SSP1- 51 2.6 (Bracegirdle et al., 2020). There is little inter-model spread in the CMIP6 future response of the 52 atmospheric westerly jet, providing high confidence in this assessment (in contrast, CMIP6 models show no 53 consistency in their future projection of easterly wind change along the Antarctic continental shelf break) 54 (Bracegirdle et al., 2020). Paleo-oceanographic evidence suggests that ACC flow through Drake Passage was 55 consistently stronger during warm intervals of the past (both during interstadials and interglacials), but with Do Not Cite, Quote or Distribute 9-34 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 relatively little change and no consensus on the sign of change in other regions (Lamy et al., 2015; Toyos et 2 al., 2020). In summary, additional evidence since the SROCC confirms that there is medium confidence that 3 the ACC has been weakly sensitive to Southern Hemisphere atmospheric jet increase in the past decades. 4 New evidence since the SROCC suggests that there is high confidence that the Southern Hemisphere 5 atmospheric jet will increase in the 21st century for all scenarios (except for SSP1-1.9 and SSP1-2.6; Section 6 4.3.3.1) with a greater increase for larger radiative forcing. An increase in westerly winds will very likely 7 force an increase of the eddy field in the ACC, and while there is medium confidence that the ACC is weakly 8 sensitive to wind change, new advances since the SROCC provide limited evidence that the ACC transport 9 will nevertheless increase in response to wind and buoyancy fluxes. 10 11 For the upper cell overturning circulation, the SROCC concluded that its transport has experienced 12 significant inter-decadal variability in response to wind forcing since the 1990s, and that there is low 13 confidence in the assessments of a long-term increase in upper ocean overturning. Consistent with the 14 SROCC, the importance of both eddy processes and winds in driving long-term change and variability have 15 been reinforced, with a potential fast wind response partially counteracted by a slower eddy response 16 (Doddridge et al., 2019; Waugh et al., 2019; Stewart et al., 2020). Eddy parameterizations affect the strength 17 of overturning, its sensitivity to winds and the ACC transport (Mak et al., 2017). Even in eddy resolving 18 simulations sub-gridscale dissipation affects the overturning and ACC (Pearson et al., 2017). In addition, 19 there has been progress in understanding the importance of Antarctic Ice Shelf meltwater and sea-ice, in 20 driving the observed changes in the near surface and in the upper overturning cell over the past decades, on 21 top of changes induced by winds and eddies (Bronselaer et al., 2020; Haumann et al., 2020; Jeong et al., 22 2020; Rye et al., 2020). In particular, increased stratification caused by increased freshwater flux to the 23 surface ocean (Section 9.2.1.3) can cause a shoaling and warming of the Circumpolar Deep Water layer, and 24 create a positive feedback enhancing basal melt of the Antarctic Ice Sheet (Section 9.4.2.1) (Bronselaer et 25 al., 2018; Golledge et al., 2019a; Schloesser et al., 2019; Sadai et al., 2020). There is medium confidence in 26 the existence of this feedback mechanism but low agreement on the magnitude of the feedback. The SROCC 27 reported that CMIP5 models project that the overall transport of upper ocean overturning cell will increase 28 by up to 20% in the 21st century, and no new studies alter that assessment. 29 30 For the lower cell overturning circulation, the SROCC assessed that a slowdown of its transport is consistent 31 with the observed decrease in volume (medium confidence) of AABW in the global ocean (Section 9.2.2.3). 32 Additional evidence since the SROCC, strengthens confidence that increased glacial meltwater flux will 33 reduce the density of bottom waters during the 21st century, eventually reaching a point where deep 34 convection will be curtailed and shelf water will become too buoyant to sink to the ocean interior, thereby 35 slowing the lower cell overturning circulation (Bronselaer et al., 2018; Golledge et al., 2019a; Lago and 36 England, 2019; Moorman et al., 2020). While such changes are consistent with the observed freshening and 37 volume decrease of the AABW layer reported in the SROCC, as discussed in Section 9.2.2.3, new 38 observation-based studies have highlighted how the lower cell overturning can episodically increase as a 39 response to climate anomalies, temporally counteracting the tendency for melt to reduce AABW formation 40 (Abrahamsen et al., 2019; Castagno et al., 2019; Gordon et al., 2020; Silvano et al., 2020). In addition, while 41 the opening of open ocean polynyas can affect the lower cell on decadal to centennial time-scales, there is 42 limited evidence and low agreement in the role of open ocean polynyas in driving observed trends of the 43 lower cell in the last decade (Section 9.2.2.3). Based on CMIP5 models, the SROCC reported with low 44 confidence that formation and export of AABW associated with the lower overturning cell will decrease in 45 the 21st century, and there is no new evidence to revisit that assessment from climate models. However, 46 additional paleo evidence from marine sediments suggest that AABW formation/ventilation was vulnerable 47 to freshwater fluxes during past interglacials (Hayes et al., 2014; Huang et al., 2020; Turney et al., 2020) and 48 that AABW formation was strongly reduced (Skinner et al., 2010; Gottschalk et al., 2016; Jaccard et al., 49 2016) or possibly totally curtailed (Huang et al., 2020) during the LGM and transient cold intervals of MIS 2 50 & 3. Specifically, sedimentary reconstructions show a transient reduction in AABW ventilation in the 51 Atlantic sector of the Southern Ocean during MIS5e, which is assessed to have been warmer than modern 52 climate (Thomas et al., 2020). However, long multi-centennial or millennial model runs under higher-than- 53 pre-industrial CO2 concentrations show that after 500 -1000 years, ventilation in the Southern Ocean 54 resumes, and even possibly overshoots with enhanced convection in the Weddell and Ross seas leading to 55 enhanced bottom water ventilation globally (Yamamoto et al., 2015; Frölicher et al., 2020). AABW Do Not Cite, Quote or Distribute 9-35 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 ventilation increased at the onset of the last deglacial transition, promoting the release of previously 2 sequestered CO2 to the atmosphere on centennial to millennial timescales (Bauska et al., 2016; Jaccard et al., 3 2016; Rae et al., 2018), concomitant with a southward shift of the SH westerly wind belt (Denton et al., 4 2010; Jaccard et al., 2016) and reduced sea-ice cover (Ferrari et al., 2014; Stein et al., 2020). In summary, 5 the combination of observational, numerical and paleoclimate evidence provides us with medium confidence 6 that the lower cell will continue decreasing in the 21st century as a result of increased basal melt from the 7 Antarctic Ice Sheet. 8 9 10 9.2.3.3 Tropical Oceans 11 12 The tropics are a tightly coupled ocean-atmosphere system with tightly interconnected basins (Cai et al., 13 2019). The zonal atmospheric Walker Circulation and the Indonesian Throughflow (ITF, Figure 9.11) are 14 key connections between the Pacific and Indian Oceans, and variations in the Walker and Hadley 15 Circulations are tightly linked to the tropical Pacific SST and currents. The tropics have a profound influence 16 on the climate system through the multiple modes of variability they host, which have widespread global 17 influence at seasonal to annual timescale (Annex IV). 18 19 The effect of tropical modes of variability on climate and their long-term changes are reviewed in detail in 20 Annex IV, while changes to the tropical ocean are assessed throughout the report and briefly summarized 21 here. Section 2.4 concludes that a sustained shift beyond multi-centennial variability has not been observed 22 for ENSO (medium confidence) and that there is limited evidence and limited agreement about the long-term 23 behaviour of other tropical modes. Section 3.7 assesses with high confidence that human influence has not 24 affected the principal tropical modes of interannual climate variability and their associated regional 25 teleconnections beyond the range of internal variability. Section 4.3.3.2 assesses with medium confidence 26 that there is no consensus from models for a systematic change in the amplitude of El Niño–Southern 27 Oscillation sea surface temperature variability over the 21st century. The related change in tropical SSTs is 28 covered in Section 9.2.1.1. The projected changes in SST have implications for marine heat wave 29 characteristics, which are assessed in Box 9.2. SST changes in the tropics are related to changes in the 30 atmospheric circulation, including surface equatorial easterly trade winds and Walker Circulation (Section 31 4.5.3.2), and the weakening Indonesian Throughflow and strengthening Agulhas Extension and leakage 32 (Section 9.2.3.4). Weakening trade winds under climate change (Vecchi and Soden, 2007) will tend to 33 decrease upwelling, along isopycnals in the eastern Pacific and diapycnal upwelling in the central Pacific and 34 thus the meridional temperature gradients that drive Tropical Instability Waves (Terada et al., 2020), along 35 with a weakening, flattening and shoaling of the tropical thermocline and equatorial undercurrent (Luo and 36 Rothstein, 2011). A weak or absent Equatorial Undercurrent (Kuntz and Schrag, 2020) and a too diffuse and 37 incorrectly sloped tropical thermocline (Zhu et al., 2020) remain issues in most CMIP6 models. In summary, 38 while future changes in tropical modes of variability remain unclear, change in atmospheric and ocean 39 circulation will drive continued change in tropical ocean temperature in the 21st century (medium 40 confidence), with part of the region experiencing drastic marine heat wave conditions (high confidence). 41 42 43 9.2.3.4 Gyres, Western Boundary Currents, and Inter-Basin Exchanges 44 45 The AR5 (Rhein et al., 2013) assessed with medium to high confidence that the North Pacific subpolar gyre, 46 the South Pacific subtropical gyre, and the subtropical cells have intensified. They also reported that the 47 North Pacific subtropical gyre had expanded since the 1990s, and that overall the changes in gyre systems 48 were likely predominantly due to interannual-to-decadal variability. The SROCC (Meredith et al., 2019) 49 complemented the AR5 assessment by reporting that the polar Beaufort Gyre in the Arctic expanded to the 50 northwest between 2003 and 2014, contemporaneous with changes in its freshwater accumulation and 51 alterations in wind forcing. Consistent with the reported change over the gyres, both the AR5 and the 52 SROCC (Bindoff et al., 2019; Collins et al., 2019) reported that Western Boundary Currents (WBCs) have 53 intensified (Figure 9.11), and expanded poleward, except for the Gulf Stream and the Kuroshio. Section 54 2.3.3.4 provides an overall assessment of gyres and WBCs including an assessment of change from 55 paleoclimate archives. Section 2.3.3.4 assesses that while WBC strength is highly variable at multidecadal Do Not Cite, Quote or Distribute 9-36 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 scale (high confidence), WBCs and subtropical gyres have shifted poleward since 1993 (medium 2 confidence), at a rate on the order of 0.04-0.1 degree per decade during 1993-2018. Figure 9.11 shows that 3 CMIP5 and CMIP6 models agree in projecting a weaker Gulf Stream and Gulf Stream Extension, while the 4 Kuroshio changes less (Sen Gupta et al., 2016). 5 6 Although the observed wind stress curl shows systematic poleward shift in each basin as a result of 7 anthropogenic warming (Section 2.3.1.4) (Chen and Wu, 2012; Wu et al., 2012; Zhai et al., 2014), which has 8 caused a systematic shift of the WBCs and subtropical gyres since 1993 (Wu et al., 2012; Yang et al., 2016b, 9 2020), the response of current strength is more complex and inconsistent across regions (Sloyan and O’Kane, 10 2015; Wang et al., 2016c; Elipot and Beal, 2018; McCarthy et al., 2018; Wang and Wu, 2018; Dong et al., 11 2019). The strength of WBCs and gyres exhibit inconsistent responses because they are not only dependent 12 on wind stress forcing, but multi-scale interaction and air-sea interaction have an important role in their long- 13 term trends and variability (Zhang et al., 2020). Observed changes in gyre circulation are dominated by 14 interannual and decadal modes of variability globally (Qiu and Chen, 2012; Melzer and Subrahmanyam, 15 2017; McCarthy et al., 2018; Hu et al., 2020). The North Atlantic subpolar gyre is strongly modulated by 16 variability associated with the NAO and AMV (Robson et al., 2016) (Annex IV). Subpolar gyre systems can 17 change abruptly due to a positive feedback between convective mixing and salinity transport (Born et al., 18 2013, 2016) and air-sea interaction (Moffa-Sánchez et al., 2014; Moreno-Chamarro et al., 2017) within the 19 gyre. In the Arctic, both the Beaufort gyre and mesoscale eddies strengthened between 2003 and 2014 20 (Armitage et al., 2017), which might be partly due to increased wind stress (Oldenburg et al., 2018b) or 21 reduced sea-ice thickness and changes in sea-ice pack morphology (van der Linden et al., 2019). Presently, 22 there is limited evidence in attributing causality to these changes for any of the proposed mechanisms. In the 23 North Pacific, there has been an increasing trend in the Alaska Gyre from 1993 to 2017 (Cummins and 24 Masson, 2018), which might be attributed to PDO (Hristova et al., 2019) (low confidence). In the Southern 25 Ocean, limited evidence indicates that the subpolar gyres respond to Southern Hemisphere atmospheric 26 modes of variability at interannual time-scale (Armitage et al., 2018; Dotto et al., 2018). 27 28 All climate models reproduce WBCs and gyres, but eddy-present or eddy-rich models (roughly 10-25 km 29 and ~10 km resolution, respectively) represent these currents more realistically than eddy-parameterized 30 models (Small et al., 2014a; Griffies et al., 2015; Chassignet et al., 2017; Hewitt et al., 2017; Roberts et al., 31 2018; Chassignet et al., 2020; Hewitt et al., 2020) (very high confidence). Compared to observations or to 32 eddy-present and eddy-rich models, the eddy-parameterized models from CMIP5 and CMIP6 simulate 33 weaker and wider WBCs as well as less realistic locations of subtropical and subpolar gyre boundaries 34 (Figure 9.11). Increased resolution not only admits mesoscale eddies, but also improves simulation of the 35 strength and position of WBCs such as the Kuroshio Current, Gulf Stream, and East Australian Current 36 (Sasaki et al., 2004; Chassignet and Marshall, 2008; Delworth et al., 2012; Yu et al., 2012; Small et al., 37 2014b; Haarsma et al., 2016; Chassignet et al., 2017, 2020; Hewitt et al., 2020) (very high confidence). 38 Improved boundary current location relates to improved recirculation regions (Jayne et al., 2009), mean path 39 and variability and existence of multiple stable paths (Qiu et al., 2005; Delman et al., 2015), air-sea fluxes 40 (Small et al., 2014a), and related coastal weather patterns (Kaspi and Schneider, 2011). The wind-current 41 feedback, implemented by considering relative velocity of currents and wind, realistically dampens 42 mesoscale eddies and WBCs, through mesoscale air-sea interaction (Ma et al., 2016; Renault et al., 2016, 43 2019), even though sub-mesoscale wind-current damping feedback is missing in these models (Zhang et al., 44 2016c) (medium confidence). As eddies potentially play a role in determining the strength of gyre 45 circulations and their low-frequency variability (Fox-Kemper and Pedlosky, 2004; Berloff et al., 2007), it is 46 expected that eddy-present and eddy-rich models will differ in their decadal variability and sensitivity to 47 changes in wind stress of gyres from eddy-parameterized models (medium confidence). Nonetheless, 48 important aspects of gyre strength depend primarily on forcing and not resolution, allowing long term 49 changes in gyre strength to be investigated with low resolution climate models (Hughes and de Cuevas, 50 2001; Yeager, 2015). 51 52 Under future scenarios RCP4.5 and RCP8.5, the AR5 (Collins et al., 2013) assessed an intensification and 53 poleward extension of the southern Hemisphere subtropical gyres in the 21st century. New evidence since the 54 AR5 further reinforce their conclusions which are now extended to all subtropical gyre systems, in both the 55 northern and southern hemispheres (Yang et al., 2016a, 2020). CMIP6 models project changes in WBCs that Do Not Cite, Quote or Distribute 9-37 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 are consistent with projected changes in the surface winds. Under strong radiative forcing, in scenario SSP5- 2 8.5, CMIP6 models project that the East Australian Current Extension and Agulhas Current Extension will 3 intensify in the 21st century, while the Gulf Stream and Brazil Current will weaken (Figure 9.11). Although 4 CMIP5/CMIP6 are limited in resolution, medium confidence is given to changes in western boundary 5 currents due to consistency across generations of climate models, including CMIP6, despite changes in 6 model structure, resolution and parameterisations. 7 8 9 [START FIGURE 9.11 HERE] 10 11 Figure 9.11: Simulated barotropic streamfunction, surface speed and major current transport in CMIP5 and 12 CMIP6. (a) Mean barotropic streamfunction (Sv) 1995-2014 and projected barotropic streamfunction 13 change (Sv, 2018-2100 vs. 1995-2014) under (b) SSP5-8.5. (d) Mean surface (0-100 m) speed (m/s) and 14 projected surface speed change (m/s, 2081-2100) versus 1995-2014 under (e) SSP5-8.5. (c, f) Median and 15 likely range of 1995-2014 and 2081-2100 transport of 3 currents with the largest transport change and 4 16 with the largest fractional change (Sen Gupta et al., 2016). (c) Deep currents: Agulhas Extension (ACx), 17 Gulf Stream (GS), Gulf Stream Extension (GSx), Tasman Leakage (TASL), East Australia Current 18 Extension (EACx), Indonesian Throughflow (ITF), and Brazil Current (BC). (f) Shallow currents: as for 19 deep but with New Guinea Current (NGC), and without ACx. No overlay indicates regions with high 20 model agreement, where ≥80% of models agree on the sign of change; diagonal lines indicate regions 21 with low model agreement, where <80% of models agree on the sign of change (see Cross-Chapter Box 22 Atlas.1 for more information). Further details on data sources and processing are available in the chapter 23 data table (Table 9.SM.9). 24 25 [END FIGURE 9.11 HERE] 26 27 28 The SROCC (Collins et al., 2019) concluded with high confidence that ITF transport from the Pacific to Indian 29 ocean has increased in the past two decades, as a result (medium confidence) of an unprecedented 30 intensification of the equatorial Pacific trade wind system. Section 2.3.3.4 assesses that there is high confidence 31 that the increase in the ITF over the past two decades is linked to multi-decadal scale variability rather than a 32 longer-term trend. Consistently, in the future, as winds change under increased radiative forcing, most models 33 project a decline of the ITF on the centennial timescale (Figure 9.11). Indeed, one of the clearest changes of 34 ocean current transport simulated by climate models is a weakening of the Indonesian Throughflow projected 35 in CMIP5 simulations under RCP4.5 and RCP8.5 scenarios (Sen Gupta et al., 2016; Stellema et al., 2019), as 36 well as in CMIP6 simulations under the SSP5-8.5 scenario (high confidence, Figure 9.11). 37 38 The SROCC reports with high confidence that the Agulhas leakage from the Indian to the Atlantic ocean has 39 increased in the past two decades (Collins et al., 2019), and there is no additional evidence since then allowing 40 to revisit this assessment (Biastoch et al., 2015; Loveday et al., 2015; Lübbecke et al., 2015). There is low 41 confidence in future projections of Agulhas leakage because most CMIP models cannot directly simulate it, 42 due to coarse resolution. However, there is medium evidence that the strength of the Southern Hemisphere 43 westerlies controls Agulhas leakage (Durgadoo et al., 2013; Biastoch et al., 2015; Loveday et al., 2015), and 44 high confidence that the strength of the Southern Hemisphere westerlies will increase under increased radiative 45 forcing except in lower warming scenarios (SSP1-1.9, SSP1.2-6; Section 4.3.3.1) (Bracegirdle et al., 2020). 46 There is also evidence that increasing Agulhas leakage is consistent with observed change of the temperature 47 and salinity structure in the Atlantic ocean, and with variability of the AMOC (Section 9.2.3.1) (Biastoch et 48 al., 2015). This range of indirect evidence provides medium confidence that the Agulhas leakage will increase 49 in the 21st century, except for the strongest mitigation scenario (Figure 9.11). 50 51 The SROCC assessed that the annual Bering Strait volume transport from the Pacific to the Arctic Ocean 52 increased from 2001–2014, consistent with an estimated increased northward heat transport of about 60% from 53 2001–2014, and an increased freshwater transport of 30±20 km3 yr–1 from 1991 to 2015 (Meredith et al., 2019). 54 Section 2.3.3.4 assesses that volume transport from the Pacific to the Arctic has increased since the 1990s from 55 0.8 Sv to 1.0 Sv over 1990-2015. Realistic representation of the Bering Strait transport in the current generation 56 of climate models is challenging because the strait is narrow compared to the resolution of climate models Do Not Cite, Quote or Distribute 9-38 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 (Clement Kinney et al., 2014; Aksenov et al., 2016). For the Atlantic to Arctic transport, Section 2.3.3.4 reports 2 that the major branches of Atlantic Water inflow across the Greenland-Scotland Ridge have remained stable, 3 with only the smaller pathway of Atlantic Water north of Iceland showing a strengthening trend during 1993- 4 2018. Section 2.3.3.4 also assesses that the Arctic outflow remained stable from the mid 1990s to the mid 5 2010s. Future changes in these currents have not yet been studied in CMIP6 models. 6 7 8 9.2.3.5 Eastern Boundary Upwelling Systems 9 10 Eastern boundary upwelling systems (EBUS) exist where trade winds draw cold and generally low-pH/low- 11 oxygen waters upward. Coastal upwelling plays a key role in supplying the food chain with nutrients, hence 12 the richness and productivity of EBUS (Bindoff et al., 2019). The SROCC (Bindoff et al., 2019) assessed 13 with high confidence that three out of the four major EBUS have experienced large-scale wind 14 intensification in the past 60 years (only the trend for the Canary current is considered uncertain). However, 15 it also emphasized that various processes can also modulate or even reverse wind trends locally (Bindoff et 16 al., 2019). Here we revisit the SROCC (Bindoff et al., 2019) assessment based on evidence showing low 17 agreement between studies that have investigated trends over past decadess of upwelling-favourable winds 18 (Varela et al., 2015). This low agreement has been related to differences in wind products, season of interest, 19 and length of the considered time series (Varela et al., 2015). Based on this, we assess that only the 20 California current system has experienced large-scale upwelling-favorable wind intensification over the 21 period 1982-2010 albeit with regional differences (García-Reyes and Largier, 2010; Seo et al., 2012). In the 22 Benguela, Canary, and Humboldt systems, large-scale, upwelling-favourable wind trends are ambiguous, 23 owing to low confidence in long-term in situ marine wind data (Cardone et al., 1990; Bakun et al., 2010) and 24 low agreement among available studies (Narayan et al., 2010; Sydeman et al., 2014; Varela et al., 2015). Our 25 assessment confirms the SROCC (Bindoff et al., 2019) in that high natural variability of EBUS and their 26 inadequate representation by most climate models gives low confidence in attribution of observed changes, 27 while anthropogenic changes are projected to emerge primarily in the second half of the 21st century (limited 28 evidence: one model and one study) (Brady et al., 2017). 29 30 Under increased radiative forcing, the SROCC (Bindoff et al., 2019) assessed that climate models project, in 31 the 21st century, a reduction of wind and upwelling intensity in EBUS at low latitudes and enhancement at 32 high latitudes under scenario RCP8.5, with an overall reduction in either upwelling intensity or extension. It 33 also highlighted that coastal warming and wind intensification may lead to variable countervailing responses 34 to upwelling intensification at local scales. Despite differences among EBUS (Wang et al., 2015a), there is 35 growing evidence since the SROCC in this pattern of change. While it has long been hypothesized that for 36 upwelling winds, change is linked to air temperature contrast between ocean and land (Bakun, 1990), this 37 hypothesis has increasingly been challenged. Changes in sea level pressure and wind fields in EBUS appear 38 to be primarily tied to those affecting subtropical highs (García‐Reyes et al., 2013). Poleward expansion of 39 the Hadley cell (Section 2.3.1.4.1) (Staten et al., 2018) and the related poleward migration of subtropical 40 highs (He et al., 2017; Cherchi et al., 2018), produce robust patterns of changes of reduced upwelling at low 41 latitude and enhanced upwelling at high latitude (Echevin et al., 2012; Belmadani et al., 2014; Bettencourt et 42 al., 2015; Rykaczewski et al., 2015; Sousa et al., 2017; Lamont et al., 2018; Sylla et al., 2019). These 43 patterns are most apparent in summer in both hemispheres. Synoptic variability of upwelling winds, 44 important to the functioning of upwelling ecosystems (García-Reyes et al., 2014), may also be affected by 45 climate change (Aguirre et al., 2019). However, coarse resolution model projections of winds in upwelling 46 regions may be more consistent than higher-resolution projections as these regions are highly sensitive to 47 resolution (Small et al., 2015). 48 49 Projected future annual cumulative upwelling wind changes at most locations and seasons remain 50 within ±10-20% of present-day values in the 21st century, even in the context of high-end emission scenarios 51 (4xCO2 or RCP8.5) (medium confidence). Changes due to wind stress curl and alongshore pressure gradients 52 do tend to agree with alongshore wind changes (Oerder et al., 2015a; Sylla et al., 2019). Direct estimation of 53 oceanic upward transport (Oyarzún and Brierley, 2019; Sylla et al., 2019) and nutrient flux into the euphotic 54 layer (Jacox et al., 2018) provide a meaningful estimator of upwelling, integrating all relevant processes, 55 including changes in wind stress curl. However, there is limited evidence from vertical velocity of climate Do Not Cite, Quote or Distribute 9-39 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 models and missing processes in coarse-resolution climate models that presently limit this approach. Change 2 in upper ocean stratification (Section 9.2.1.3) is projected to increase confinement of upwelling vertical 3 velocities to near the ocean surface (Oerder et al., 2015a; Oyarzún and Brierley, 2019) (high confidence). 4 5 In summary, the SROCC and we conclude that the California current system has experienced some 6 upwelling-favourable wind intensification since the 1980s (high confidence), while low agreement among 7 reported wind changes in the Benguela, Canary, and Humboldt systems prevents a similar assessment. As in 8 the SROCC, there is low confidence in attribution of observed changes to anthropogenic or natural causes. 9 New evidence reinforces our confidence in the SROCC assessment that under increased radiative forcing, 10 EBUS winds will change with a dipole spatial pattern within each EBUS of reduction (weaker and/or 11 shorter) at low latitude, and enhancement (stronger and/or longer) at high latitude (high confidence). There is 12 medium confidence that, across all scenarios, upwelling wind changes in EBUS will remain moderate in the 13 21st century, within ±10-20% from present-day values. 14 15 16 9.2.3.6 Coastal Systems and Marginal Seas 17 18 Beyond the world’s coastlines lie the shoreline, shallow estuaries, continental shelves, and deeper fjords and 19 slopes, where depths increase rapidly from the shelves to the deep ocean floor. It is more difficult to 20 transport fluid across the shelf-break or slope than along (Brink, 2016), and estuaries and shelves have 21 complex circulations and mixing leading to indirect connections between the inner shelves and coastlines 22 and offshore conditions. Coastal processes link to both large-scale metrics of climate and regional effects, 23 from changing rivers and estuaries, melt and runoff to deep water, to how changes offshore affect regional 24 and coastal conditions. 25 26 Shelf-deep ocean exchanges involve eddying, tidal, or turbulent motions and small-scale topography such as 27 submarine canyons; high resolution observations and models are needed to capture these effects (Greenberg 28 et al., 2007; Capet et al., 2008; Allen and Durrieu de Madron, 2009; Colas et al., 2012; Trotta et al., 2017). 29 Example coastal processes that introduce uncertainty into large-scale projections are exchange of CDW 30 across the Antarctic shelf-break, which affects AABW formation and Antarctic ice shelf-ocean interaction 31 (Sections 9.2.2.3, 9.2.3.2) (Stewart and Thompson, 2013, 2015), river and estuarine plumes and their 32 responses to water level and hydrology change (Banas et al., 2009; Sun et al., 2017), fjord dynamics linked 33 to glacial outflows (Straneo and Cenedese, 2015; Torsvik et al., 2019), and changing formation of water 34 masses in marginal seas (Kim et al., 2001; Greene and Pershing, 2007; Giorgi and Lionello, 2008; Renner et 35 al., 2009). Downscaling projections to the local level allows process detail (Foreman et al., 2013; Mathis and 36 Pohlmann, 2014; Meier, 2015; Tinker et al., 2016). Some processes can only be simulated when coastal 37 models are forced by larger-scale models of the atmosphere, cryosphere, or hydrosphere (Seo et al., 2007, 38 2008; Somot et al., 2008; Oerder et al., 2015b; Renault et al., 2016; Zhang et al., 2016a; Wåhlin et al., 2020), 39 including the addition of tides (Janeković and Powell, 2012; Timko et al., 2013; Tinker et al., 2015; 40 Pickering et al., 2017; Hausmann et al., 2020). Due to coastal process complexity and small scale, linking the 41 effects of coastal ocean changes to global ocean changes requires high resolution modelling (Holt et al., 42 2017, 2018), two-way nesting, or local mesh refinement (Fringer et al., 2006; Zhang and Baptista, 2008; 43 Mason et al., 2010; Dietrich et al., 2012; Hellmer et al., 2012; Ringler et al., 2013; Wang et al., 2014b; Zängl 44 et al., 2015; Zhang et al., 2016b; Soto-Navarro et al., 2020). Coarse climate models and even HighResMIP 45 models do not represent some coastal phenomena such as cross-shelf exchanges and sub-mesoscale eddies 46 which require 1km or finer resolution. Thus, there is low confidence in projecting centennial scale coastal 47 climate change where regional downscaling or refinement is lacking. There is high confidence in the ability 48 of regional coupled models to improve coastal climate change process understanding and provide regional 49 information (Section 12.4), but many sites globally await such projections. 50 51 52 53 54 Do Not Cite, Quote or Distribute 9-40 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 9.2.4 Steric and dynamic sea-level change 2 3 9.2.4.1 Global mean thermosteric sea-level change 4 5 Changes in globally averaged ocean heat content (OHC) cause global mean thermosteric sea-level (GMTSL) 6 change (Box 9.1). The observed increased OHC for 1971-2018 of 325 to 546 ZJ (very likely range, Section 7 7.2, Box 7.2) has led to a GMTSL rise of 0.03 to 0.06 m out of a total GMSL of 0.07 to 0.15 m (very likely 8 range, Section 2.3.3.3, Table 2.7, Table 9.5, Cross-Chapter Box 9.1). 9 10 Projections of GMTSL rise in the AR5 (Church et al., 2013a) and the SROCC (Oppenheimer et al., 2019) 11 were derived from the CMIP5 ensemble, after removing drift estimated based on pre-industrial control 12 simulations. Differences between removing a linear and a quadratic drift are small (Hermans et al. 2021) 13 (Hobbs et al., 2016b; Hermans et al., 2021). These prior assessments filled in projections for models that did 14 not provide GMTSL rise for all scenarios, by calculating the heat content of the climate system from global 15 surface air temperature and net radiative flux, then converting this to GMTSL rise using each model’s 16 diagnosed expansion efficiency coefficient. In the AR5, the associated uncertainties were derived by 17 assuming a normal distribution, with the 5th-95th percentile CMIP5 ensemble range taken as the likely range 18 (±1 standard deviation). 19 20 In this report, global surface air temperature projections are not derived directly from the CMIP6 ensemble 21 (Box 4.1). Therefore, in order to produce projections of OHC and GMTSL rise that are consistent with the 22 report’s assessment of equilibrium climate sensitivity and transient climate response (Section 7.5.2.2), this 23 chapter employs a two-layer energy budget emulator (Supplementary Materials 7.SM.2, 9.SM.4.3). Since the 24 AR5, climate model emulators have been increasingly used to predict GMTSL (Kostov et al., 2014; Palmer 25 et al., 2018, 2020; Nauels et al., 2019) (Cross-Chapter Box 7.1). The expansion efficiency coefficient that 26 relates GMTSL and OHC for the two-layer emulator has a mean and standard deviation of 0.113 ± 0.013 27 m/YJ (Supplementary Material 9.SM.4.3). This approach yields a likely thermosteric contribution between 28 1995 to 2014 and 2100 that represents a minimal change from the AR5 and the SROCC (Table 9.8). The 29 two-layer emulator GMTSL projected median and 17th-83rd percentile, or likely, range is 0.12 (0.09-0.15) m 30 for SSP1-1.9, 0.14 (0.11-0.18) m for SSP1-2.6, 0.20 (0.16-0.24) m for SSP2-4.5, 0.25 (0.21-0.30) m for 31 SSP3-7.0, and 0.30 (0.24-0.36) m for SSP5-8.5 by 2100 (Section 9.6.3.2; Tables 9.1, 9.8,9.9). The two-layer 32 model heat content increases slightly faster than that of the total depth CMIP6 ensemble, which is related to 33 its role in the assessed energy balance (Section 7.SM.2), but with a similar ensemble spread (Table 9.1). 34 Projecting the likely factor by which 1995-2014 to 2081-2100 ocean heat content change exceeds change 35 over 1971 to 2018 in CMIP6 yields 3 to 5 for SSP1-2.6, 4 to 6 for SSP2-4.5, 5 to 7 for SSP3-7.0, and 5 to 8 36 for SSP5-8.5. The two-layer model likely equivalents are 2 to 3 for SSP1-2.6, 3 to 4 for SSP2-4.5, 4 to 5 for 37 SSP3-7.0, and 4 to 6 for SSP5-8.5. 38 39 For reconstructions, the expansion efficiency coefficient is required for the conversion between ocean 40 temperature and steric sea level over a specific time scale. Combining the assessed sea level and energy data 41 over 1995 to 2014 (drawn from the analysis in Cross-Chapter Box 9.1) results in a coefficient of 0.1210 ± 42 0.0014 m/YJ, or 0.6607 ± 0.0076 m/ºC in terms of mean ocean temperature. The two-layer emulator 43 assessment used in AR6 results in 0.113 ± 0.013 m/YJ, or 0.617 ± 0.071 m/ºC (Appendices 7.SM.2, 9.SM.4). 44 Both of these estimates are in line with an independent estimate of 0.70 m/ºC (Hieronymus, 2019) and other 45 estimates, e.g., 0.116 ± 0.011 m/YJ (Kuhlbrodt and Gregory, 2012), but are significantly larger than the 46 temperature to sea level conversion used in the AR5 (0.42 m/ºC based on SST and the estimated range from 47 (Levermann et al., 2013)). The expansion coefficient is not fixed across models nor in time, as it varies 48 depending on which water masses are storing the added heat and the commitment time scale (Hallberg et al., 49 2013). For paleoclimate, a scaling for sea surface temperature (0.6 m/ºC) or GSAT (see Cross-Chapter Box 50 2.3) can be estimated, but mean ocean temperature is in phase with steric sea-level change while sea surface 51 temperatures are not (Shakun et al., 2012; Tierney et al., 2020) (Figure 9.9). Thus, while conversions 52 between OHC, mean ocean temperature and GMTSL across applications are within uncertainty ranges 53 (medium confidence, Table 9.1), little consistency is found when correlating these variables to SST or GSAT 54 which may vary independently. 55 Do Not Cite, Quote or Distribute 9-41 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Short-lived climate forcers (Sections 6.3, 6.6.3) are associated with a sea-level commitment, due to an ocean 2 heat content and mean ocean temperature response that lasts substantially longer than their atmospheric 3 forcing and SST response, although not as long as the sea-level commitment associated with CO2 emissions 4 (Sections 9.2.1.1, 4.4.4). For example, Zickfeld et al., (2017) find that about 70% of the thermosteric sea- 5 level rise associated with methane forcing would persist 100 years after the elimination of methane 6 emissions and 40% would persist for over 500 years. 7 8 In summary, consistent relationships between OHC (Section 9.2.2.1), mean ocean temperature and GMTSL 9 are found using two-layer emulators, CMIP6 models, and modern and paleo observations to provide medium 10 confidence in the 0.113 ± 0.013 m/YJ, or 0.617 ± 0.071 m/ºC likely ranges of assessed conversion values. It 11 is possible to estimate relationships between SST or GSAT change and GMTSL rise, but conversions are not 12 generally applicable and depend on time scale and application. 13 14 15 [START TABLE 9.1 HERE] 16 17 Table 9.1: Projected contributions to median and 17-83% (parentheses) and 5-95% (square brackets) ranges of 18 thermosteric sea level from AR5 (Church et al., 2013), CMIP6 (Jevrejeva et al., 2020; Hermans et al., 19 2021) and the two-layer energy balance model (described in Sections 7.SM.2, 9.SM.4 and Box 4.1) 20 averaged over 2081-2100, with respect to a baseline of 1995-2014. Note that AR5 and SROCC interpret 21 5-95% range as the likely range, while in this table square brackets are used for consistency. 22 Study RCP2.6/SSP1-2.6 RCP4.5/SSP2-4.5 RCP8.5/SSP5-8.5 IPCC AR5 and 0.13 [0.09 – 0.17] m 0.18 [0.13 - 0.22] m 0.26 [0.20 – 0.32] m SROCC GMTSL (Oppenheimer et al., 2019) (Church et al., 2013a) CMIP6 5-95% 0.14 [0.08 – 0.17] m 0.18 [0.11 – 0.23] m 0.26 [0.17 – 0.33] m GMTSL (Hermans et al. 2021) CMIP6 5-95% – 0.19 [0.13 – 0.24] m 0.27 [0.19 – 0.35] m GMTSL (Jevrejeva et al., 2020) Assessed GMTSL 0.13 (0.11 – 0.16) 0.17 (0.14 – 0.21) 0.25 (0.20 – 0.30) based on two-layer [0.09 – 0.19] m [0.12 – 0.25] m [0.18 – 0.35] m model 17-83% and 5- 95% (Sections 7.SM.2, 9.SM.4) Total OHC 17-83% 1.18 (0.99 – 1.42) [0.86 – 1.56 (1.33 – 1.86) [1.19 – 2.23 (1.92 – 2.64) [1.71 – and 5-95% from 1.65] YJ 2.12] YJ 3.00] YJ assessed two-layer model (Sections 7.SM.2, 9.SM.4) 0-2000m OHC 17- 1.35 (1.08 – 1.67) [0.90 – 1.89 (1.60 – 2.29) [1.28 – 83% and 5-95% 1.06 (0.80 – 1.31) [0.66 – 1.84] YJ 2.58] YJ from CMIP6 (Figure 1.64] YJ 9.6) 23 24 [END TABLE 9.1] 25 26 27 9.2.4.2 Ocean dynamic sea-level change 28 29 Projections of ocean dynamic sea-level change (Box 9.1) on multiannual timescales resemble the patterns of 30 steric sea-level change in the open ocean (Figures 9.11 and 9.12) (Lowe and Gregory, 2006; Pardaens et al., Do Not Cite, Quote or Distribute 9-42 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 2011; Couldrey et al., 2020). On shorter timescales, especially in extratropical coastal areas, there may be an 2 important barotropic component (also called bottom pressure change) due mostly to changes in wind-driven 3 circulation and eddies apparent in the variance of ocean dynamic sea level (Figure 9.12, (Roberts et al., 4 2016; Hughes et al., 2018). This component is highly sensitive to ocean model resolution (Chassignet et al., 5 2020). Steric sea-level change is associated with local changes in temperature and salinity, which come 6 about through changes in surface fluxes of heat and freshwater (Section 9.2.1.2) and through redistribution of 7 existing water masses by changed ocean circulation and mixing processes (Figure 9.12, Sections 9.2.2.1, 8 9.2.3). Redistribution of water masses often involves anticorrelated thermosteric and halosteric changes 9 (Figure 9.12), especially in the Atlantic (Pardaens et al., 2011; Bouttes et al., 2014; Durack et al., 2014; 10 Griffies et al., 2014; Han et al., 2017). 11 12 13 [START FIGURE 9.12 HERE] 14 15 Figure 9.12: (a-f) CMIP6 multi-model mean projected change contributions to relative sea level change in (a,d) 16 steric sea level anomaly, (b, e) thermosteric sea level anomaly, and (c, f) halosteric sea level 17 anomaly between 1995-2014 and 2081-2100 using a method that does not require a reference level 18 (Landerer et al., 2007). Global mean change has been removed from these figures, consistent with the 19 methods in Sections 9.6.3 and 9.SM.4.3 and the definitions of (Gregory et al., 2019). See Figure 9.27 for 20 GMSL. (g-i) Standard deviation of ocean dynamic sea-level change from (g) Aviso observations (10 day 21 highpass filter), (h) 5-day mean of high-resolution OMIP-2 models forced with observed fluxes, and (i) 5- 22 day mean of low-resolution OMIP-2 models which are comparable in resolution to the models in (a-f). 23 No overlay indicates regions with high model agreement, where ≥80% of models agree on the sign of 24 change; diagonal lines indicate regions with low model agreement, where <80% of models agree on the 25 sign of change (see Cross-Chapter Box Atlas.1 for more information). Further details on data sources and 26 processing are available in the chapter data table (Table 9.SM.9). 27 28 [END FIGURE 9.12 HERE] 29 30 Ocean dynamic sea-level change is strongly affected by internal variability (Section 9.6.1.4), partly from 31 interannual to decadal coupled atmosphere-ocean modes of variability via wind-driven redistribution (Annex 32 IV; Griffies et al., 2014; Han et al., 2017) and partly from intrinsic ocean variability, particularly in higher 33 resolution simulations (such as HighResMIP), which statistically resemble observations, even on short 34 timescales (Figure 9.12, Griffies et al., 2014; Sérazin et al., 2016; Llovel et al., 2018; Chassignet et al., 35 2020). High-resolution simulations are not used in relative sea level projections (Section 9.6.3) due to the 36 limited range of forcing scenarios. The most marked feature of long-term regional sea-level change in the 37 continuous satellite altimetry record, beginning in 1992, is the east-west dipole in the Pacific Ocean (rising 38 more rapidly in the east, see also Section 9.6.1.3), which persisted until 2015 and can be explained by 39 anomalously strong trade winds (Merrifield et al., 2012; England et al., 2014; Griffies et al., 2014; Takahashi 40 and Watanabe, 2016; Han et al., 2017) together with associated changes in surface heat flux (Piecuch et al., 41 2019). The most notable features of sub-annual variability in altimetry are eddies and tides, which are 42 directly simulated only in high resolution models (Haigh et al., 2019; Chassignet et al., 2020). 43 44 Projections of the pattern and amplitude of regional ocean dynamic sea-level change in CMIP6 and previous 45 model generations show a large model spread, of a similar size to the geographical spread (Figure 9.12). The 46 model spread derives from model dependence of changes both in surface fluxes (Section 9.2.1.2) and in the 47 ocean response (Section 9.2.2). The spread is similar in CMIP6 and CMIP5, and is largest in regions with 48 large projected variations in ensemble-mean ocean dynamic sea-level change (Lyu et al., 2020a), such as the 49 Southern Ocean dipole with an ocean dynamic sea-level rise north of the ACC and a fall to the south, the 50 Atlantic dipole with a sea-level rise north of 40ºN and a fall in 20-40ºN, the north-west Pacific dipole, and 51 the large sea-level rise in the Arctic (Church et al., 2013a; Slangen et al., 2014b; Bilbao et al., 2015; Slangen 52 et al., 2015; Gregory et al., 2016; Chen et al., 2019a; Couldrey et al., 2020; Lyu et al., 2020a). Patterns of 53 change are consistent between model simulations and observations (medium confidence). The major model 54 ensemble-mean features resemble thermosteric sea level change, as expected from altered input of heat to the 55 ocean without changing circulation, while model spread results from the diversity in redistribution of the 56 heat content of the unperturbed ocean (Section 9.2.2.1; (Bouttes and Gregory, 2014; Gregory et al., 2016; Do Not Cite, Quote or Distribute 9-43 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Huber and Zanna, 2017; Couldrey et al., 2020; Lyu et al., 2020b; Todd et al., 2020)). 2 3 The Southern Ocean meridional dipole is driven by a northward advection of excess heat (from changes in 4 surface fluxes) by the wind-driven circulation followed by subduction or diffusive uptake in mid-latitudes, 5 northward redistribution of existing heat by the strengthening of that circulation, and the meridional contrast 6 in thermal expansivity due to its temperature-dependence (Armour et al., 2016; Gregory et al., 2016; 7 Couldrey et al., 2020; Lyu et al., 2020b; Todd et al., 2020). 8 9 The positive Arctic ocean dynamic sea-level change is driven by increased freshwater input (Couldrey et al., 10 2020). The north-west Pacific dipole is driven by the intensification of the Kuroshio current in response to 11 reduced heat loss and in some models to wind stress change (Chen et al., 2019a; Couldrey et al., 2020). 12 13 The North Atlantic sea-level change dipole is forced by a reduction in heat loss from the ocean north of 40ºN 14 (i.e., net heat uptake), which in all Earth system models leads to a weakening of the AMOC, although the 15 magnitude has a large model spread (Gregory et al., 2016; Huber and Zanna, 2017, Section 9.2.3.1). The 16 reduced northward transport of warm, salty water (Section 9.2.2) causes further ocean dynamic sea-level 17 change, whose details are model-dependent. North of 40ºN, this redistribution leads to a sea-level rise, 18 predominantly halosteric, reinforcing the thermosteric effect of heat uptake (Couldrey et al., 2020). 19 Comparison of observed Atlantic ocean heat content for 1955-2017 with a reconstruction assuming no 20 change in circulation indicates that the thermosteric sea-level change resulting from southward redistribution 21 of heat may be detectable (Zanna et al., 2019a). This redistribution causes a tendency for SST cooling north 22 of 40ºN and anomalous heat input from the atmosphere, and thus a positive feedback on AMOC weakening 23 (Winton et al., 2013; Gregory et al., 2016; Couldrey et al., 2020; Todd et al., 2020). Many climate and ocean 24 models agree that the AMOC weakening is associated with pronounced thermosteric sea-level rise along the 25 American coast around 40ºN (Figures 9.12, 9.26), leading to a relatively large ocean dynamic sea-level rise 26 in this region (Yin, 2012; Bouttes et al., 2014; Slangen et al., 2014a; Little et al., 2019; Lyu et al., 2020a). 27 28 In summary, ocean dynamic sea-level change involves changes to temperature and salinity and responses of 29 currents to changing forcing, with significant variability driven by unforced oceanic variability. Projections 30 of dynamic sea-level variability require fully three-dimensional ocean models and only high-resolution ocean 31 models are statistically consistent on short timescales with satellite altimeter observations (very high 32 confidence). 33 34 35 9.3 Sea ice 36 37 9.3.1 Arctic Sea Ice 38 39 9.3.1.1 Arctic Sea-Ice Coverage 40 41 The observed decrease of Arctic sea-ice area is a key indicator of large-scale climate change (Section 42 2.3.2.1.1, Cross-Chapter Box 2.2). The SROCC (Meredith et al., 2019) assesses that sea-ice extent, which is 43 the total area of all grid cells with at least 15% sea-ice concentration, has declined since 1979 in each month 44 of the year (very high confidence). In contrast to the SROCC, we assess changes in sea-ice area (the actual 45 area of the ocean covered by sea ice) rather than sea-ice extent, because sea-ice area is geophysically more 46 relevant and not grid-dependent (Notz, 2014; Ivanova et al., 2016; Notz et al., 2016; Notz and SIMIP 47 Community, 2020). Arctic sea-ice area is calculated based on measurements by passive microwave satellite 48 sensors that provide near-continuous measurements of gridded, pan-Arctic sea-ice concentration from 1979 49 onwards. Irreducible uncertainties in the conversion of thermal microwave brightness temperature to sea-ice 50 concentration and choices in algorithm design cause uncertainties in observed Arctic sea-ice area, which are, 51 though, far smaller than the observed sea-ice loss (e.g., Comiso et al., 2017b; Niederdrenk and Notz, 2018; 52 Alekseeva et al., 2019; Kern et al., 2019; Meier and Stewart, 2019). Sea-ice area has decreased from 1979 to 53 the present in every month of the year (very high confidence, Figure 9.13). The absolute and the relative ice 54 losses are highest in late summer-early autumn (high confidence, Figure 9.13). Averaged over the decade 55 2010-2019, monthly-average Arctic sea-ice area in August, September and October has been around 2 Do Not Cite, Quote or Distribute 9-44 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 million km² (or about 25%) smaller than that during 1979-1988 (high confidence, Figure 9.13). 2 3 4 [START FIGURE 9.13 HERE] 5 6 Figure 9.13: Arctic sea-ice historical records and CMIP6 projections.Left: Absolute anomaly of monthly-mean 7 Arctic sea-ice area during the period 1979 to 2019 relative to the average monthly-mean Arctic sea-ice 8 area during the period 1979 to 2008. Right: Sea-ice concentration in the Arctic for March and September, 9 which usually are the months of maximum and minimum sea-ice area, respectively. First column: 10 Satellite-retrieved mean sea-ice concentration during the decade 1979-1988. Second column: Satellite- 11 retrieved mean sea-ice concentration during the decade 2010-2019. Third column: Absolute change in 12 sea-ice concentration between these two decades, with grid lines indicating non-significant differences. 13 Fourth column: number of available CMIP6 models that simulate a mean sea-ice concentration above 15 14 % for the decade 2045-2054. The average observational record of sea-ice area is derived from the UHH 15 sea-ice area product (Doerr et al., 2021), based on the average sea-ice concentration of OSISAF/CCI 16 (OSI-450 for 1979-2015, OSI-430b for 2016-2019)(Lavergne et al., 2019), NASA Team (version 1, 17 1979-2019)(Cavalieri et al., 1996) and Bootstrap (version 3, 1979-2019)(Comiso, 2017) that is also used 18 for the figure panels showing observed sea-ice concentration. Further details on data sources and 19 processing are available in the chapter data table (Table 9.SM.9). 20 21 [END FIGURE 9.13 HERE] 22 23 24 The SROCC discussed the regional distribution of Arctic sea-ice loss and their findings remain valid for the 25 updated time series covering 2019 (Figure 9.13). Sea-ice loss in winter is strongest in the Barents Sea, while 26 summer losses occur primarily at the summer sea-ice region margins, in particular in the East Siberian, 27 Chukchi, Kara and Beaufort Seas (Frey et al., 2015; Chen et al., 2016; Onarheim et al., 2018; Peng and 28 Meier, 2018; Maksym, 2019). In the Bering Sea, expanding winter sea-ice cover was observed until 2017 29 (Frey et al., 2015; Onarheim et al., 2018; Peng and Meier, 2018), but a marked reduction in sea-ice 30 concentration has occurred since then (Stabeno and Bell, 2019)(high confidence). 31 32 With respect to seasonal changes in the sea-ice cover, the winter sea-ice loss causes a decrease in the average 33 sea-ice age and fraction of multi-year ice as assessed by the SROCC (very high confidence), and also of the 34 ocean area covered intermittently by sea ice (Bliss et al., 2019). In contrast, the seasonal ice zone (covered 35 by sea ice in winter but not in summer) has expanded regionally (Bliss et al., 2019) and over the whole 36 Arctic (Steele and Ermold, 2015), because the loss of summer sea-ice area is larger than the loss of winter 37 sea-ice area. Arctic sea ice retreat includes an earlier onset of surface melt in spring and a later freeze up in 38 fall, lengthening the open-water season in the seasonal sea-ice zone (Stroeve and Notz, 2018). However, 39 there is low agreement in quantification of regional trends of melt and freeze onset between different 40 observational products (Bliss et al., 2017; Smith and Jahn, 2019). 41 42 Reconstructions of Arctic sea-ice coverage put the satellite period changes into centennial context. Direct 43 observational data coverage (Walsh et al., 2017) and model reconstructions (Brennan et al., 2020) warrant 44 high confidence that the low Arctic sea-ice area of summer 2012 is unprecedented since 1850, and that the 45 summer sea-ice loss is significant in all Arctic regions except for the Central Arctic (Cai et al., 2021). Direct 46 wintertime observational data coverage before 1953 is too sparse to reliably assess Arctic sea-ice area. Since 47 1953, the years 2015 to 2018 had the four lowest values of maximum Arctic sea-ice area, which usually 48 occurs in March (Figure 2.20) (high confidence). Reconstructions of Arctic sea-ice area before 1850 remain 49 sparse, and as in the SROCC, there remains medium confidence that the current sea-ice levels in late summer 50 are unique during the past 1 kyr (Kinnard et al., 2011; De Vernal et al., 2013b). (Section 2.3.2.1.1) 51 52 The observed fluctuations and trends of the Arctic sea-ice cover arise from a combination of changes in 53 natural external forcing and anthropogenic forcing, internal variability and internal feedbacks (e.g., Notz and 54 Stroeve, 2018; Halloran et al., 2020). New paleo-proxy techniques indicate regional sea-ice changes over 55 epochs and millennia and allow possible drivers to be assessed. Biomarker IP25 (Belt et al., 2007) together 56 with other sedimentary biomarkers (Belt, 2018) provides local temporal information on seasonal sea-ice Do Not Cite, Quote or Distribute 9-45 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 coverage, permanent sea-ice coverage and ice-free waters with occasional ambiguous contrasting results 2 (Belt, 2019). These records and other proposed paleo proxies including bromine in ice cores (Spolaor et al., 3 2016), dinocyst assemblages (e.g., De Vernal et al., 2013b) and driftwood (e.g., Funder et al., 2011) provide 4 evidence of sea-ice fluctuations that exceed internal variability (high confidence). 5 6 The inferred sea-ice fluctuations over millennia can be related to Northern-hemisphere temperature evolution 7 and give rise to Arctic-wide fluctuations in sea-ice coverage in the paleo record (Section 2.3.2.1.1). On a 8 regional scale, fluctuations include decreased sea-ice cover during the Allerød warm period (14.7-12.9 ka) in 9 the Laptev (Hörner et al., 2016) and Bering Sea (Méheust et al., 2018); an extensive sea-ice cover during the 10 Younger Dryas (~12 ka) in the Bering (Méheust et al., 2018), Kara (Hörner et al., 2018), Laptev (Hörner et 11 al., 2016) and Barents (Belt et al., 2015) Seas and at the Yermak Plateau (Kremer et al., 2018); little sea ice 12 during the early Holocene, when Northern hemisphere summer insolation was higher than today (8000 to 13 9000 years before present), in the North Icelandic Shelf area (Cabedo-Sanz et al., 2016; Xiao et al., 2017), 14 Sea of Okhotsk (Lo et al., 2018), Canadian Arctic (Spolaor et al., 2016), Barents (Berben et al., 2017), 15 Bering (Méheust et al., 2018), and Chukchi (Stein et al., 2017) Seas, at the Yermak Plateau (Kremer et al., 16 2018) and north of Greenland (Funder et al., 2011); increasing sea-ice cover throughout much of the middle 17 and late Holocene around Svalbard (Knies et al., 2017), in the North Icelandic Shelf area (Cabedo-Sanz et 18 al., 2016; Harning et al., 2019; Halloran et al., 2020), north of Greenland (Funder et al., 2011), and in the 19 Western Greenland (Kolling et al., 2018), Barents (Belt et al., 2015; Berben et al., 2017), Chukchi (De 20 Vernal et al., 2013a; Stein et al., 2017) and Laptev (Hörner et al., 2016) Seas. The consistent, Arctic-wide 21 changes give high confidence in millennial-scale co-variability of the sea-ice cover with temperature 22 fluctuation. 23 24 The SROCC assessed that approximately half of the satellite-observed Arctic summer sea ice loss is driven 25 by increased concentrations of atmospheric greenhouse gases (medium confidence). Recent attribution 26 studies now allow the strengthened assessment that it is very likely that more than half of the observed Arctic 27 sea-ice loss in summer is anthropogenic (Section 3.4.1.1). This assessment is confirmed by process-based 28 analyses of Arctic sea-ice loss not assessed by the SROCC. Similar to the paleo record, the satellite record of 29 Arctic sea-ice area from 1979 onwards is strongly linearly correlated with global mean temperature on 30 decadal and longer time scales (Figure 9.14a,e) (e.g., Gregory et al., 2002; Rosenblum and Eisenman, 2017) . 31 The correlation holds across all months with R2 ranging from 0.61 to 0.81 (Niederdrenk and Notz, 2018). 32 However, in contrast to paleo-records, sea-ice fluctuations during the satellite period are only weakly 33 correlated with Northern Hemisphere insolation (Notz and Marotzke, 2012); modern Northern Hemisphere 34 sea-ice area is more strongly correlated with atmospheric CO2 concentration (Johannessen, 2008; Notz and 35 Marotzke, 2012) and cumulative anthropogenic CO2 emissions (Figure 9.14b,f) (Zickfeld et al., 2012; 36 Herrington and Zickfeld, 2014; Notz and Stroeve, 2016). R2 values of the correlation between sea-ice area 37 and cumulative CO2 emissions range across all months from 0.76 to 0.92 (Stroeve and Notz, 2018). In 38 summary, there is high confidence that satellite-observed Arctic sea-ice area is strongly correlated with 39 global mean temperature, CO2 concentration and cumulative anthropogenic CO2 emissions. 40 41 In addition to changes in the external forcing, internal variability substantially affects Arctic sea ice, 42 evidenced from both paleo records (e.g., (Chan et al., 2017; Hörner et al., 2017; Kolling et al., 2018)) and 43 satellites after 1979 (e.g., Notz and Stroeve, 2018; Roberts et al., 2020) (high confidence). Most of the 44 internal variability on annual time scales is related to atmospheric temperature fluctuations, for example 45 linked to cyclone activities (Wernli and Papritz, 2018; Olonscheck et al., 2019), while multidecadal internal 46 variability is primarily related to changes in oceanic heat transport (Zhang, 2015; Halloran et al., 2020). 47 These mechanisms are represented in current climate models (Olonscheck et al., 2019; Halloran et al., 2020), 48 but the resulting internal variability of September sea-ice area in CMIP5 and CMIP6 models, as given by the 49 ensemble mean standard deviation s SIA,Sep=0.5 million km² (Olonscheck and Notz, 2017; Notz and SIMIP 50 Community, 2020), exceeds the estimated internal variability for the period 1850 to 1979 from both 51 reanalyses (s SIA,Sep =0.3 million km2) and direct observational reconstructions (sSIA,Sep =0.2 million km2) 52 (Brennan et al., 2020) (medium confidence because of limited reliability of longer-term sea-ice 53 reconstructions). Internal variability has been estimated to have contributed 30 to 50% of the observed Arctic 54 summer sea-ice loss since 1979 (Kay et al., 2011; Stroeve et al., 2012; Ding et al., 2017, 2019; England et 55 al., 2019). However, this estimate from models might be biased towards internal over forced variability Do Not Cite, Quote or Distribute 9-46 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 because of the models’ high internal variability and because the CMIP5 simulated September sea-ice 2 sensitivity to forcing is lower than observed, even if internal variability is taken into account (Notz and 3 Stroeve, 2016; Rosenblum and Eisenman, 2017). Most CMIP6 models fail to simulate the observed 4 sensitivity of sea-ice loss to CO2 emissions (as a proxy for time) and to temperature simultaneously. 5 However, they better capture the observed sensitivity of sea-ice loss to CO2 emissions than CMIP5 models 6 (Section 3.4.1, Figure 9.14h; (Notz and SIMIP Community, 2020)). 7 8 The SROCC examined the different atmospheric and oceanic processes that caused the observed sea-ice loss, 9 with recent studies providing new evidence for the importance of variations in air temperature (Olonscheck 10 et al., 2019; Dahlke et al., 2020), wind patterns (Graham et al., 2019), oceanic heat flux (Docquier et al., 11 2021) and riverine heat influx (Park et al., 2020). As in the SROCC, the relative contribution of each 12 physical cause to the sea-ice loss cannot be robustly quantified because of disagreement among models 13 (Burgard and Notz, 2017), sparse observations and limited understanding of the variation of each factor with 14 global mean temperature. This is addressed by new diagnostics available from CMIP6 simulations, which 15 now allow for more detailed analyses of the drivers of sea-ice loss at a process level (Keen et al., 2021). 16 17 In examining temperature thresholds for the loss of Arctic summer sea ice, the SR1.5 (Hoegh-Guldberg et 18 al., 2018) and the SROCC assess that a reduction of September-mean sea-ice area to below 1 million km2, 19 practically a sea-ice-free Arctic Ocean, is more probable for a global mean warming of 2C compared to 20 global mean warming of 1.5C (high confidence). Analyses of CMIP6 simulations (Notz and SIMIP 21 Community, 2020) confirm this result, as they show that on decadal and longer time scales, Arctic summer 22 sea ice area will remain highly correlated with global mean temperature until the summer sea ice has 23 vanished (Figure 9.14a,e). Quantitatively, existing studies (Screen and Williamson, 2017; Jahn, 2018; Ridley 24 and Blockley, 2018; Sigmond et al., 2018; Notz and SIMIP Community, 2020) additionally show that for a 25 warming between 1.5 and 2 ˚C, the Arctic will only be practically sea-ice free in September in some years, 26 while at 3 ˚C warming the Arctic is practically sea-ice free in September in most years, with longer 27 practically sea-ice-free periods at higher warming levels (medium confidence). However, because of the 28 CMIP5 and CMIP6 models’ generally too low sensitivity of sea-ice loss to global warming, there is only low 29 confidence regarding the specific warming level at which the Arctic Ocean first becomes practically sea-ice 30 free (Notz and SIMIP Community, 2020). (Section 4.3.2.1) 31 32 33 [START FIGURE 9.14 HERE] 34 35 Figure 9.14: Monthly mean March (a-d) and September (e-h) sea-ice area as a function of global surface air 36 temperature (GSAT) anomaly (a,e); cumulative anthropogenic CO2 emissions (b,f); year (c,g) in 37 CMIP6 model simulations (shading, ensemble mean as bold line) and in observations (black dots). 38 Panels d and h show the sensitivity of sea-ice loss to anthropogenic CO2 emissions as a function of the 39 modelled sensitivity of GSAT to anthropogenic CO2 emissions. In panels d and h, the black dot denotes 40 the observed sensitivity, while the shading around it denotes internal variability as inferred from CMIP6 41 simulations (after Notz and SIMIP Community, 2020). Further details on data sources and processing are 42 available in the chapter data table (Table 9.SM.9). 43 44 [END FIGURE 9.14 HERE] 45 46 47 In contrast, CMIP6 models capture the observed sensitivity of Arctic sea ice area to cumulative 48 anthropogenic CO2 emissions well, providing high confidence that the Arctic Ocean will likely become 49 practically sea-ice free in the September mean for the first time for future CO2 emissions of less than 1000 Gt 50 and before the year 2050 in all SSP scenarios (Notz and SIMIP Community, 2020). This new assessment is 51 consistent with an observation-based projection of a practically sea-ice free Arctic Ocean in September for 52 additional anthropogenic CO2 emissions of 800± 330 GtCO2 beyond the year 2018 (Notz and Stroeve, 2018; 53 Stroeve and Notz, 2018). This estimate may, however, be too high due to neglecting possible future 54 reduction in atmospheric aerosol load that would cause additional warming (Gagné et al., 2015a; Wang et al., 55 2018) and is subject to the same constraints as the carbon budget analysis for global mean temperature (see 56 section 5.5 for details). Based on CMIP6 simulations, it is very likely that the Arctic Ocean will remain sea- Do Not Cite, Quote or Distribute 9-47 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 ice covered in winter in all scenarios throughout this century (Sections 4.3.2, 4.4.2). 2 3 There is indication that CMIP6 simulations of Arctic sea ice have improved relative to CMIP5 (Section 4 3.4.1.1), but detailed evaluation studies exist mainly for CMIP5 models. These studies found that CMIP5 5 model projections and reanalyses show a large spread of simulated regional Arctic sea-ice concentration 6 (Laliberté et al., 2016; Chevallier et al., 2017), which remains true for CMIP6 models (Shu et al., 2020; Wei 7 et al., 2020). In addition, both CMIP5 and CMIP6 models show a large spread in the simulated seasonal 8 cycle of Arctic sea-ice area, with too high a sea-ice area in March in the ensemble mean (Notz and SIMIP 9 Community, 2020). CMIP5 models also have been found to have difficulty simulating realistic landfast sea 10 ice (Laliberté et al., 2018). These findings imply that both CMIP5 and CMIP6 models do not realistically 11 capture the regional and seasonal processes governing observed Arctic sea-ice evolution, causing low 12 confidence in the models’ projections of future regional sea-ice evolution, including updated projections for 13 shipping routes across the Northern Sea Route and Northwest Passage (Wei et al., 2020). 14 15 CMIP5 models also have issues with capturing the seasonal cycle of observed changes in Arctic sea-ice drift 16 speed, which affects their simulation of regional sea-ice concentration patterns. Direct measurements of 17 Arctic sea ice from drift buoys and satellites show that drift speed of Arctic sea ice has increased over the 18 satellite period in all seasons (e.g., Rampal et al., 2009; Docquier et al., 2017). In summer, CMIP5 models 19 show a slowdown of Arctic sea-ice drift rather than the observed acceleration (Tandon et al., 2018). In 20 winter, CMIP5 models generally capture the observed acceleration of Arctic drift speed. The drift 21 acceleration is primarily caused by the decrease in concentration and thickness, both in the observational 22 record (Rampal et al., 2009; Spreen et al., 2011; Olason and Notz, 2014; Docquier et al., 2017) and, for 23 winter, in CMIP5 models (Tandon et al., 2018). Changes in wind speed are less important for the observed 24 large-scale changes (Spreen et al., 2011; Vihma et al., 2012; Olason and Notz, 2014; Docquier et al., 2017; 25 Tandon et al., 2018). In summary, there is high confidence that Arctic sea-ice drift has accelerated because of 26 the decrease in sea ice concentration and thickness. 27 28 The SR1.5 assessed with high confidence that there is no hysteresis in the loss of Arctic summer sea ice. In 29 addition, there is no tipping point or critical threshold in global mean temperature beyond which the loss of 30 summer sea ice becomes self-accelerating and irreversible (high confidence). This is because stabilizing 31 feedbacks during winter related to increased heat loss through thin ice and thin snow, and increased emission 32 of longwave radiation from open water, dominate over the amplifying ice-albedo feedback (e.g., Eisenman, 33 2012; Wagner and Eisenman, 2015; Notz and Stroeve, 2018) (see section 7.4.2 for details on the individual 34 feedbacks). Observed and modelled Arctic summer sea ice and global mean temperature are linked with little 35 temporal delay, and the summer sea-ice loss is reversible on decadal time scales (Armour et al., 2011; Ridley 36 et al., 2012; Li et al., 2013; Jahn, 2018). The loss of winter sea ice is reversible as well, but the loss of winter 37 sea-ice area per degree of warming in CMIP5 and CMIP6 projections increases as the ice retreats from the 38 continental shore lines, because these limit the possible areal fluctuations (e.g., Bathiany et al., 2016, 2020; 39 Meccia et al., 2020) (high confidence) (Section 4.3.2.1). 40 41 42 9.3.1.2 Arctic Sea-Ice volume and thickness 43 44 The SROCC assessed with very high confidence that Arctic sea ice has become thinner over the satellite 45 period from 1979 onwards, and this assessment is confirmed for the updated time series (section 2.3.2.1.1). 46 Sea-ice area has also decreased substantially over this period (section 9.3.1.1), leading to the assessment that 47 Arctic sea-ice volume has also decreased with very high confidence over the satellite period since 1979. 48 There is, however, only low confidence in quantitative estimates of the sea-ice volume loss over this period 49 because of a lack of reliable, long-term, pan-Arctic observations and substantial spread in available 50 reanalyses (Chevallier et al., 2017). Current best estimates from reanalyses suggest a reduction of September 51 Arctic sea ice volume of 55 to 65 % over the period 1979 to 2010, and of about 72 % over the period 1979 to 52 2016, with the latter deemed a conservative estimate (Schweiger et al., 2019). 53 54 For the more recent past, ice-thickness can be directly estimated from satellite estimates of sea-ice freeboard 55 (Kwok and Cunningham, 2015; Kwok, 2018). Based on these retrievals, there is medium confidence that Do Not Cite, Quote or Distribute 9-48 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Arctic sea-ice volume has decreased since 2003. There is low confidence in the amount of decrease over this 2 period and over the CryoSat-2 period from 2011 onwards primarily because of snow-induced uncertainties in 3 the retrieval algorithms, the shortness of the record, and the small identified trend (e.g., (Bunzel et al., 2018; 4 Petty et al., 2018, 2020)). 5 6 Observations of regional changes in sea-ice thickness vary in quality. Analysis of submarine data in the 7 central Arctic Ocean suggests that sea ice there has thinned by about 75 cm compared to the mid-1970s 8 (Section 2.3.2.1.1). For smaller regions, data are too sparse to allow for quantitative estimates of long-term 9 trends (King et al., 2017; Rösel et al., 2018), but a clear thinning signal over 10 to 20 years has been found 10 for sea ice in Fram Strait (Spreen et al., 2020), north of Canada (Haas et al., 2017) and for landfast ice in 11 Kongsfjorden/Svalbard (Pavlova et al., 2019). CMIP5 models and reanalyses fail to capture the observed 12 distribution (Stroeve et al., 2014; Shu et al., 2015) and evolution (Chevallier et al., 2017) of Arctic sea-ice 13 thickness. Most CMIP6 models do not capture the observed spatial distribution of sea-ice thickness 14 realistically (Wei et al., 2020). This leads to low confidence in estimates of thickness from reanalyses and 15 from CMIP5 and CMIP6 models, and in these models’ projections of sea-ice volume. 16 17 18 9.3.2 Antarctic Sea Ice 19 20 9.3.2.1 Antarctic sea-ice coverage 21 22 The SROCC (Meredith et al., 2019) assessed that there was no significant trend in annual mean Antarctic 23 sea-ice area over the period of reliable satellite retrievals starting in 1979 (high confidence). The updated 24 time series is consistent with this assessment. It includes a maximum sea-ice area in 2014, a substantial 25 decline from then until the minimum sea-ice area in 2017, and an increase in sea-ice area since then 26 (Schlosser et al., 2018; Maksym, 2019; Parkinson, 2019) (Figure 9.15, Figure 2.20). As assessed in Section 27 2.3.2.1.2, the possible significance of the increase in mean Antarctic sea-ice area over the shorter period 28 1979 to 2014 (Figure 2.20) (Simmonds, 2015; Comiso et al., 2017a) is unclear. This is because of 29 observational uncertainty (see section 9.3.1.1), large year-to-year fluctuations in all months (Figure 9.15), 30 and limited understanding of the processes and reliability of year-to-year correlation of Antarctic sea-ice area 31 (Yuan et al., 2017). 32 33 As assessed by the SROCC, the evolution of mean Antarctic sea-ice area is the result of opposing regional 34 trends (high confidence), with slightly decreasing sea-ice cover during the period 1979 to 2019 in the 35 Amundsen Sea and the Bellingshausen Sea, particularly during summer, and slightly increasing sea-ice cover 36 in the eastern parts of the Weddell Sea and the Ross Sea (Figure 9.15). With the exception of the Ross Sea, 37 these trends are not significant considering the large variability of the time series (Yuan et al., 2017). 38 39 The SROCC assessed that the regional trends are closely related to meridional wind trends (high confidence). 40 This is the case as the regional trends in the maximum northward extent of the ice cover (Figure 9.15) are 41 determined by the balance between the northward advection of the ice that is formed in polynyas near the 42 continental margin, and the lateral and subsurface melting through oceanic heat fluxes. The advection of the 43 sea ice is strongly correlated with winds and cyclones (Schemm, 2018; Vichi et al., 2019; Alberello et al., 44 2020). Accordingly, the increasing sea-ice area in the Ross Sea can be linked to a strengthening of the 45 Amundsen Sea low (e.g., (Holland et al., 2017b, 2018)), while other regional sea-ice trends in the austral fall 46 can be linked to changes in westerly winds, cyclone activity and the Southern Annular Mode (SAM) in 47 summer and spring (Doddridge and Marshall, 2017; Holland et al., 2017a; Schemm, 2018). In addition to the 48 wind-driven changes, increased near-surface ocean stratification (Section 9.2.1.3) has contributed to the 49 observed increase in sea-ice coverage (e.g., (Purich et al., 2018; Zhang et al., 2019b)) as it tends to cool the 50 surface ocean (Sections 9.2.1.1, 9.2.3.2). The changes in stratification result partly from surface freshening 51 (De Lavergne et al., 2014) (associated with increased northward sea-ice advection (Haumann et al., 2020) 52 and/or melting of the Antarctic ice sheet (e.g., (Haumann et al., 2020; Jeong et al., 2020; Mackie et al., 53 2020)) (medium confidence)) and are amplified by local ice-ocean feedbacks (Goosse and Zunz, 2014; 54 Lecomte et al., 2017; Goosse et al., 2018). In the Amundsen Sea, strong ice-shelf melting can cause local 55 sea-ice melt next to the ice-shelf front by entraining warm Circumpolar Deep Water to the ice-shelf cavity Do Not Cite, Quote or Distribute 9-49 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 and surface ocean (medium confidence) (Sections 9.2.3.2, 9.4.2.2) (Jourdain et al., 2017; Merino et al., 2 2018). It has also been suggested that the observed regional increase in sea ice coverage since 1979 results 3 from a long-term Southern Ocean surface cooling trend (e.g., (Kusahara et al., 2019; Jeong et al., 2020)) but 4 the importance of this mechanism for the observed sea-ice evolution is unclear owing to intricate feedbacks 5 between sea-ice change and surface cooling (Haumann et al., 2020). The importance of changing wave 6 activity (Section 9.6.4.2; Kohout et al., 2014; Bennetts et al., 2017; Roach et al., 2018b) on sea ice is unclear 7 due to limited process understanding. In summary, there is high confidence that regional Antarctic trends are 8 primarily caused by changes in sea-ice drift and decay, with medium confidence in a dominating role of 9 changing wind pattern. The precise relative contribution of individual drivers remains uncertain because of 10 limited observations, disagreement between models, unresolved processes, and temporal and spatial remote 11 linkages caused by sea-ice drift (Section 9.2.3.2, (Pope et al., 2017)). 12 13 Recent research has confirmed the SROCC assessment of atmospheric and oceanic drivers of the sea-ice 14 decline from 2014 to 2017, which can be linked to changes in both subsurface ocean heat flux (Meehl et al., 15 2019; Purich and England, 2019) and atmospheric circulation, with the latter partly related to teleconnections 16 with the tropics (Meehl et al., 2019; Purich and England, 2019; Wang et al., 2019a). In the Weddell Sea, 17 these changes caused in 2017 the re-emergence of the largest polynya over the Maud Rise since the 1970s 18 (Campbell et al., 2019; Jena et al., 2019; Turner et al., 2020) (Section 9.2.3.2). 19 20 21 [START FIGURE 9.15 HERE] 22 23 Figure 9.15: Antarctic sea-ice historical records and CMIP6 projections. Left: Absolute anomaly of observed 24 monthly-mean Antarctic sea-ice area during the period 1979 to 2019 relative to the average monthly- 25 mean Antarctic sea-ice area during the period 1979 to 2008. Right: Sea-ice coverage in the Antarctic as 26 given by the average of the three most widely used satellite-based estimates for September and February, 27 which usually are the months of maximum and minimum sea-ice coverage, respectively. First column: 28 Mean sea-ice coverage during the decade 1979-1988. Second column: Mean sea-ice coverage during the 29 decade 2010-2019. Third column: Absolute change in sea-ice concentration between these two decades, 30 with grid lines indicating non-significant differences. Fourth column: number of available CMIP6 models 31 that simulate a mean sea-ice concentration above 15 % for the decade 2045-2054. The average 32 observational record of sea-ice area is derived from the UHH sea-ice area product (Doerr et al., 2021), 33 based on the average sea-ice concentration of OSISAF/CCI (OSI-450 for 1979-2015, OSI-430b for 2016- 34 2019)(Lavergne et al., 2019), NASA Team (version 1, 1979-2019)(Cavalieri et al., 1996) and Bootstrap 35 (version 3, 1979-2019)(Comiso, 2017) that is also used for the figure panels showing observed sea-ice 36 concentration. Further details on data sources and processing are available in the chapter data table (Table 37 9.SM.9). 38 39 [END FIGURE 9.15 HERE] 40 41 42 The AR5 (Collins et al., 2013) and the SROCC found low confidence in future projections of Antarctic sea 43 ice. This includes the projected mitigation of the sea-ice loss by stratospheric ozone recovery (Smith et al., 44 2012) and by an increased freshwater input from melting of the Antarctic ice sheet (Bronselaer et al., 2018). 45 Compared to the interannual variability during the satellite record from 1979 onwards, models simulate too 46 much variability both in CMIP5 (Zunz et al., 2013) and in CMIP6 (Roach et al., 2020). The seasonal cycle in 47 sea-ice coverage is misrepresented in most CMIP5 (e.g., (Holmes et al., 2019)) and CMIP6 models (Roach et 48 al., 2020), but the multi-model mean seasonal cycle in CMIP5 and CMIP6 agrees well with observations 49 (Shu et al., 2015; Roach et al., 2020). Most CMIP5 models do not realistically simulate the evolution of 50 Antarctic sea-ice volume (Shu et al., 2015) and consistently overestimate the amount of low concentration 51 sea ice and underestimate the amount of high concentration sea ice (Roach et al., 2018a). CMIP6 models, in 52 contrast, simulate a more realistic distribution of regional sea-ice coverage (Roach et al., 2020). Most CMIP5 53 models poorly represent Antarctic sea-ice drift (e.g., (Schroeter et al., 2018; Holmes et al., 2019)), affecting 54 simulated historical trends, with models that simulate a strong sea-ice motion showing more variability in 55 sea-ice coverage than models with weaker sea-ice motion (Schroeter et al., 2018). Owing to limited 56 agreement between model simulations and observations, limited reliable observations on a process level and Do Not Cite, Quote or Distribute 9-50 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 a lack of process understanding of the substantial spread in CMIP5 and CMIP6 model simulations, there 2 remains low confidence in existing future projections of Antarctic sea-ice evolution. 3 4 The discrepancy between the modelled and observed evolution of Antarctic sea ice has been related by the 5 SROCC to deficiencies in stratification, freshening by ice shelf melt water, clouds, and other wind and ocean 6 driven processes. Recent studies highlight the possible mis-representation of freshwater fluxes from ice 7 shelves (Jeong et al., 2020), and the possible effect of the low resolution of most models (Sidorenko et al., 8 2019), even though lower-resolution models are in principle capable of a realistic simulation of the seasonal 9 sea-ice budgets in the Southern Ocean (Holmes et al., 2019). The relative importance of these possible 10 reasons for model-shortcomings remains unclear (see section 3.4.1.2 for details). 11 12 The analysis and understanding of the long-term evolution of the Antarctic sea-ice cover is hindered by the 13 scarcity of observational records before the satellite period and the scarcity of paleo records (see section 14 2.3.2.1.2 for further details). Such long records are particularly relevant given that the Southern Ocean 15 response to external forcing takes longer than the length of the available direct observational record (Goosse 16 and Renssen, 2001; Armour et al., 2016). There is only limited evidence for large-scale decadal fluctuations 17 in sea-ice coverage caused by large-scale temperature and wind forcing. Sparse direct pre-satellite 18 observations suggest a decrease in sea-ice coverage from the 1950s to the 1970s (Fan et al., 2014). Paleo- 19 proxy data indicate that, on multi-decadal to multi-centennial time scales, sea-ice coverage of the Southern 20 Ocean follows large-scale temperature trends (e.g., (Crosta et al., 2018; Chadwick et al., 2020; Lamping et 21 al., 2020)), for example linked to fluctuations in the El Niño Southern Oscillation and Southern Annular 22 Mode (Crosta et al., 2021), and that during the Last Glacial Maximum Antarctic sea ice extended to about 23 the polar front latitude in most regions during winter, whereas the extent during summer is less well 24 understood (e.g., (Benz et al., 2016; Xiao et al., 2016; Nair et al., 2019)). 25 26 Regionally, proxy data from ice cores consistently indicate that the increase of sea-ice area in the Ross Sea 27 and the decrease of sea-ice area in the Bellingshausen Sea are part of longer centennial trends and exceed 28 internal variability on multi-decadal time-scales (e.g., (Thomas et al., 2019; Tesi et al., 2020)) (medium 29 confidence). These centennial trends are consistent with simulations from CMIP5 models (Hobbs et al., 30 2016a; Jones et al., 2016b; Kimura et al., 2017). 31 32 There is low confidence in the attribution of the observed changes in Antarctic sea-ice area (Section 3.4.1.2). 33 Based on the available evidence, the lack of a negative trend of Antarctic sea-ice area despite substantial 34 global warming in recent decades has been attributed to internal variability in analyses of the observational 35 record (Meier et al., 2013; Gallaher et al., 2014; Gagné et al., 2015b), reconstructions from early 36 observations (Fan et al., 2014; Edinburgh and Day, 2016) and proxy data (Hobbs et al., 2016a), and model 37 simulations (Turner et al., 2013; Zunz et al., 2013; Zhang et al., 2019c). Nonetheless, without accurate 38 simulations of observed changes, the possible contribution of anthropogenic forcing to the regional changes 39 in sea-ice area remains unclear (Hosking et al., 2013; Turner et al., 2013; Haumann et al., 2014; Zhang et al., 40 2019c) 41 42 The attribution of the observed trends in atmospheric and oceanic forcing is also uncertain because of limited 43 observational records and discrepancies between modelled and observed evolution of the sea-ice cover. More 44 specifically, there is contrasting evidence for a direct role of stratospheric ozone depletion on the observed 45 changes in atmospheric circulation (Haumann et al., 2014; England et al., 2016; Landrum et al., 2017). In 46 contrast, there is high confidence that multi-decadal variations in the tropical Pacific and in the Atlantic 47 affect the Amundsen Sea low (Li et al., 2014; Kwok et al., 2016; Meehl et al., 2016; Purich et al., 2016; 48 Simpkins et al., 2016), while other modes of climate variability (Annex IV) affect, for example, Southern 49 Ocean cyclone activity (Simpkins et al., 2012; Cerrone et al., 2017; Schemm, 2018). 50 51 52 9.3.2.2 Antarctic sea-ice thickness 53 54 The SROCC assessed that observations are too sparse to reliably estimate long-term trends in Antarctic sea- 55 ice thickness. This remains true, and only qualitative statements on prevailing thicknesses are possible. Data Do Not Cite, Quote or Distribute 9-51 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 from ICESat-1 laser altimetry (Kurtz and Markus, 2012), from Operation IceBridge (Kwok and Kacimi, 2 2018), and long-term ship-board observations collected in the ASPeCt data set (Worby et al., 2008) suggest 3 that sea ice thicker than 1 m prevails in regions of multi-year ice along the eastern coast of the Antarctic 4 Peninsula in the Weddell Sea, in the high-latitude embayment of the Weddell Sea, and along the coast of the 5 Amundsen Sea, with remaining regions dominated by thinner first year sea ice (high confidence). Regional 6 patterns in ice thickness are affected by large snow deposition and resulting snow-ice formation (Massom et 7 al., 2001; Maksym and Markus, 2008), and deformation, ridging, and rafting that regionally cause formation 8 of very thick sea ice (Massom et al., 2006; Williams et al., 2015a). In addition, near ice shelves a sub-ice 9 platelet layer from supercooled water can significantly increase sea-ice thickness (Hoppmann et al., 2020; 10 Haas et al., 2021). Regarding snow thickness, observations are too sparse in space and time to reliably 11 estimate changes across Southern Ocean sea ice (Webster et al., 2018). 12 13 There is low confidence in the long-term trend of Antarctic sea-ice thickness. Both ASPeCt and ICESat-1 14 measurements are biased low in regions with thick ice (Kern and Spreen, 2015), compared to results from 15 reanalyses (Massonnet et al., 2013; Haumann et al., 2016) and observations with autonomous vehicles under 16 sea ice (Williams et al., 2015a). Estimates of sea-ice thickness from Cryosat-2 do not substantially reduce 17 uncertainty, primarily because of the unknown snow thickness and radar scattering above the snow–ice 18 interface (Bunzel et al., 2018; Kwok and Kacimi, 2018; Kacimi and Kwok, 2020). Isolated in-situ time series 19 show no clear long-term trend in landfast ice thickness in the Weddell Sea (Arndt et al., 2020). Reanalyses 20 suggest overall increasing sea-ice thickness and volume between 1980 and 2010 (Holland et al., 2014; 21 Zhang, 2014; Massonnet et al., 2015), while CMIP5 (Shu et al., 2015; Schroeter et al., 2018) and CMIP6 22 models simulate a decrease in Antarctic sea-ice volume over the historical period. Because of this 23 discrepancy, and the unclear reliability of the reanalyses (Uotila et al., 2019), there is low confidence in 24 CMIP5 and CMIP6 simulated future Antarctic sea-ice thickness. 25 26 27 9.4 Ice Sheets 28 29 9.4.1 Greenland Ice Sheet 30 31 9.4.1.1 Recent observed changes 32 33 In this section we present regional mass change time series for the Greenland Ice Sheet and assess the 34 different processes causing the increase in mass loss. The vast increase in observational products from 35 various platforms (e.g, GRACE, PROMICE, ESA-CCI, NASA MEaSUREs) provide a consistent and clear 36 picture of a shrinking Greenland Ice Sheet (Colgan et al., 2019; Mottram et al., 2019; Mouginot et al., 2019; 37 King et al., 2020; Mankoff et al., 2020; Moon et al., 2020; Sasgen et al., 2020; The IMBIE Team, 2020; 38 Velicogna et al., 2020). Section 2.3.2.4.1 provides an updated estimate of the total Greenland Ice Sheet mass 39 change in a global context (Figure 2.24). A paleo perspective on Greenland Ice Sheet evolution is presented 40 in Section 9.6.2 with estimated ice sheet extent at different times shown in Figure 9.17. 41 42 For the 20th century, the SROCC (Meredith et al., 2019) presented one reconstruction for 1900-1983 and 43 estimated mass change for the Greenland Ice Sheet and its peripheral glaciers for the period 1901-1990. 44 Since the SROCC, a comprehensive new study has extended the satellite record back to 1972 (Mouginot et 45 al., 2019,Figure 9.16). The rate of change of ice sheet mass was positive (i.e., it gained mass) in 1972-1980 46 (47±21 Gt yr-1) and then negative (i.e., it lost mass) (-51±17 Gt yr-1 and -41±17 Gt yr-1) in 1980-1990 and 47 1990-2000, respectively. Other ice discharge time series starting in 1985 (King et al., 2018, 2020, Mankoff et 48 al., 2019, 2020) agree with (Mouginot et al., 2019, Figure 9.16). There is limited evidence of temporally and 49 spatially heterogeneous Greenland outlet glacier evolution during 20th century (Lea et al., 2014; Lüthi et al., 50 2016; Andresen et al., 2017; Khan et al., 2020; Vermassen et al., 2020). Historical photographs (Khan et al., 51 2020) show large mass losses of Jakobshavn and Kangerlussuaq glaciers in West Greenland from 1880 until 52 the 1940s, exceeding their 21st century mass loss, whereas the Helheim Glacier in East Greenland remained 53 stable, gained mass in the 1990s then rapidly lost mass after 2000. Together, these 3 large outlet glaciers, 54 draining ~12% of the ice sheet surface area, have lost 22±3 Gt yr-1 in the period 1880-2012 (Khan et al., 55 2020). Overall, these studies provide a variable picture of the Greenland Ice Sheet mass change in the 20 th Do Not Cite, Quote or Distribute 9-52 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 century. The updated mass loss of Greenland Ice Sheet including peripheral glaciers for period 1901-1990 is 2 120 [70–170] Gt yr-1 (see Table 9.5 and Figures 9.16, 9.17). 3 4 Post-1992, the SROCC stated that it is extremely likely that the rate of mass change of Greenland Ice Sheet 5 was more negative during 2012-2016 than during 1992-2001, with very high confidence that summer melting 6 has increased since the 1990s to a level unprecedented over at least the last 350 years. Since the SROCC, the 7 updated synthesis of satellite observations by the Ice Sheet Mass Balance Intercomparison Exercise (The 8 IMBIE Team, 2020) and the GRACE Follow-On (GRACE-FO) Mission (Abich et al., 2019; Kornfeld et al., 9 2019), have confirmed the mass change record and the record has been extended to 2020 (The IMBIE team, 10 2021) as presented in 2.3.2.4.. The Greenland Ice Sheet lost 4890 [4140–5640] Gt of ice between 1992 and 11 2020, causing sea level to rise by 13.5 [11.4–15.6] mm (The IMBIE Team, 2021) (Section 2.3.2.4.1; Figure 12 9.16, Table 9.5). The IMBIE Team (2020) estimates are consistent with other post-AR5 reviews (Figure 13 9.17, Table 9.SM.1) (Bamber et al., 2018a; Cazenave et al., 2018; Mouginot et al., 2019; Slater et al., 2021). 14 Recent GRACE-FO data (Sasgen et al., 2020; Velicogna et al., 2020) show that after two cold summers in 15 2017 and 2018, with relatively moderate mass change of about -100 Gt yr-1, the 2019 mass change (-532 ± 16 58 Gt yr-1) was the largest annual mass loss in the record. The high agreement across a variety of methods 17 confirms the SROCC and Chapter 2 assessments. The mass-loss rate was on average 39 [–3 to 80] Gt yr-1 18 over the period 1992–1999, 175 [131 to 220] Gt yr-1 over the period 2000–2009 and 243 [197 to 290] Gt yr-1 19 over the period 2010–2019 (see Table 9.SM.1). 20 21 22 [START FIGURE 9.16 HERE] 23 24 Figure 9.16: Mass changes and mass change rates for Greenland and Antarctic ice sheet regions. (Upper Left) 25 Time series of mass changes in Greenland for each of the major drainage basins shown in the inset figure 26 (Bamber et al., 2018b; Mouginot et al., 2019) for the periods 1972 – 2018 and 1992-2018. (Upper Right) 27 Time series of mass changes for three portions of Antarctica (Bamber et al., 2018b) for the period 1992 – 28 2018. (Lower rows) Estimates of mass change rates of surface mass balance, discharge and mass balance 29 in seven Greenland regions (Bamber et al., 2018b; Mankoff et al., 2019; Mouginot et al., 2019; King et 30 al., 2020). Estimates of mass change rates of surface mass balance, discharge and mass balance for three 31 regions of Antarctica (Bamber et al., 2018b; The IMBIE Team et al., 2018; Rignot et al., 2019). Further 32 details on data sources and processing are available in the chapter data table (Table 9.SM.9). 33 34 [END FIGURE 9.16 HERE] 35 36 37 The SROCC assessed with high confidence that surface mass balance (SMB), rather than discharge, has 38 started to dominate the mass loss of the Greenland Ice Sheet (due to increased surface melting and runoff), 39 increasing from 42% of the total mass loss for 2000–2005 to 68% for 2009–2012. While these estimates 40 have been confirmed since the SROCC (Mouginot et al., 2019), the new longer record, as well as further 41 comprehensive studies (Khan et al., 2015; Colgan et al., 2019; Mottram et al., 2019; The IMBIE Team, 42 2020) and detailed discharge records (King et al., 2020; Mankoff et al., 2020) reveal a more complex picture 43 than the continuous trajectory this statement may have implied. Discharge was relatively constant from 44 1972-1999, varying by around 6% for the whole ice sheet, while SMB varied by over a factor of two 45 interannually, leading to either mass gain or loss in a given year (Figure 9.16). During 2000-2005, the rate of 46 discharge increased by 18%, then remained fairly constant again (increasing by 6% from 2006-2018). After 47 2000, SMB decreased more rapidly than discharge increased. In summary, the consistent temporal pattern in 48 these longer datasets leads to high confidence that the Greenland Ice Sheet mass losses are increasingly 49 dominated by SMB, but there is high confidence that mass loss varies strongly, due to large interannual 50 variability in SMB. 51 52 On a regional scale, the surface elevation is lowering in all regions, and widespread terminus and calving 53 front retreats have been observed (with no glaciers advancing) (Mottram et al., 2019; Moon et al., 2020). The 54 largest mass losses have occurred along the west coast and in southeast Greenland (Figure 9.16), 55 concentrated at a few major outlet glaciers (Mouginot et al., 2019; Khan et al., 2020). This regional pattern is 56 consistent with independent Global Navigation Satellite System (GNSS) observations from the Greenland Do Not Cite, Quote or Distribute 9-53 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 GPS network which show elastic bedrock uplift of tens of centimetres between 2007-2019 as a result of 2 ongoing ice mass loss (Bevis et al., 2019). The regional time series (Figures 9.16, Atlas.30) show that SMB 3 has been gradually decreasing in all regions while the increase in discharge in the southeast, central east, 4 northwest and central west has been linked to retreating tidewater glaciers (Figure 9.16). In summary, the 5 detailed regional records show an increase in mass loss in all regions after the 1980s, caused by both 6 increases in discharge and decreases in SMB (high confidence), although the timing and patterns vary 7 between regions. The largest mass loss occurred in the northwest and the southeast of Greenland (high 8 confidence). 9 10 11 [START FIGURE 9.17 HERE] 12 13 Figure 9.17: Greenland Ice Sheet cumulative mass change and equivalent sea level contribution. (a) A p-box 14 (Section 9.6.3.2) based estimate of the range of values of paleo Greenland ice sheet mass and sea level 15 equivalents relative to present day and the median over all central estimates (Simpson et al., 2009; Argus 16 and Peltier, 2010; Fyke et al., 2011; Robinson et al., 2011; Colville et al., 2011; Dolan et al., 2011; Born 17 and Nisancioglu, 2012; Miller et al., 2012b; Helsen et al., 2013; Nick et al., 2013; Quiquet et al., 2013; 18 Stone et al., 2013; Dahl-Jensen et al., 2013; Lecavalier et al., 2014; Robinson and Goelzer, 2014; 19 Colleoni et al., 2014; Koenig et al., 2015; Peltier et al., 2015; Calov et al., 2015; Stuhne and Peltier, 2015; 20 Vizcaino et al., 2015; Calov et al., 2018; Dutton et al., 2015; Goelzer et al., 2016; Khan et al., 2016; Yau 21 et al., 2016; de Boer et al., 2017; Simms et al., 2019); and (b) cumulative mass loss (and sea level 22 equivalent) from 1972 (Mouginot et al., 2019) and 1992 (Bamber et al., 2018b; The IMBIE Team, 2019), 23 the estimated mass loss from 1840 (Box and Colgan, 2013; Kjeldsen et al., 2015) indicated with a shaded 24 box and projections from ISMIP6 by 2100 under RCP8.5/SSP5-8.5 and RCP2.6/SSP1-2.6 scenarios (thin 25 lines from (Goelzer et al., 2020; Edwards et al., 2021; Payne et al., 2021) and likely and very likely range 26 of the ISMIP6 emulation (shades and bold line (Edwards et al., 2021)) are shown as a timeseries. 27 Schematic interpretations of individual reconstructions (Lecavalier et al., 2014; Goelzer et al., 2016; 28 Berends et al., 2019) of the spatial extent of the Greenland ice sheet are shown for the (c) mid-Pliocene 29 Warm Period, (d) the Last Interglacial and (e) the Last Glacial Maximum: grey shading shows extent of 30 grounded ice. Maps of mean elevation changes (f) 2010-2017 derived from CryoSat 2 radar altimetry 31 (Bamber et al., 2018b) and (g) ISMIP6 model mean (2093-2100) projected changes for the MIROC5 32 climate model under the RCP8.5 scenario (Goelzer et al., 2020). Further details on data sources and 33 processing are available in the chapter data table (Table 9.SM.9). 34 35 [END FIGURE 9.17 HERE] 36 37 38 The SROCC stated with high confidence that variability in large-scale atmospheric circulation is an 39 important driver of short-term SMB changes for the Greenland Ice Sheet. This effect of atmospheric 40 circulation variability on both precipitation and melt rates (and the SROCC assessment) is confirmed by 41 more recent publications (Välisuo et al., 2018; Zhang et al., 2019a; Velicogna et al., 2020). The strong mass 42 loss in 2019 (Cullather et al., 2020; Hanna et al., 2020; Tedesco and Fettweis, 2020) was driven by highly 43 anomalous atmospheric circulation patterns, both on daily (Cullather et al., 2020) and seasonal timescales 44 (Tedesco and Fettweis, 2020). Although surface melt is anticorrelated with the summer North Atlantic 45 Oscillation index (Välisuo et al., 2018; Ruan et al., 2019; Sherman et al., 2020), especially in West 46 Greenland (Bevis et al., 2019), Greenland Ice Sheet melt is more strongly correlated with the Greenland 47 Blocking Index (Hanna et al., 2016, 2018) than with the summer North Atlantic Oscillation index (Huai et 48 al., 2020). 49 50 The SROCC did not assess the role of cloud changes in detail. Studies since the AR5 have shown that higher 51 incident shortwave radiation in conjunction with reduced cloud cover leads to increased melt rates, 52 particularly over the low albedo ablation zone in the southern part of the Greenland Ice Sheet (Hofer et al., 53 2017; Niwano et al., 2019; Ruan et al., 2019). Conversely, an increase in cloud cover over the high-albedo 54 central parts of the ice sheet, leading to higher downwelling longwave radiation, was shown to lead either to 55 increased melt (Bennartz et al., 2013) or reduced refreezing of meltwater (van Tricht et al., 2016). The 56 elevation dependence of the cloud radiative effect and its control on surface meltwater generation and 57 refreezing (Wang et al., 2019c; Hahn et al., 2020) can induce a spatially consistent response of the integrated Do Not Cite, Quote or Distribute 9-54 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Greenland Ice Sheet melt to dominant patterns of cloudand atmospheric variability. The short- and longwave 2 radiation effects on surface melt by clouds have been shown to compensate for each other during events of 3 strong atmospheric rivers and the increase in melt is caused by increased sensible heat fluxes during such 4 events (Mattingly et al., 2020). In summary, there is medium confidence that cloud cover changes are an 5 important driver of the increasing melt rates in the southern and western part of the Greenland Ice Sheet. 6 7 The SROCC stated with high confidence that positive albedo feedbacks contributed substantially to the post- 8 1990s Greenland Ice Sheet melt increase. Several, mostly positive, feedbacks involving surface albedo 9 operate on ice sheets (e.g., Fyke et al., 2018). Melt amplification by the observed increase of bare ice 10 exposure through snowline migration to higher parts of the ice sheet since 2000 (Shimada et al., 2016; Ryan 11 et al., 2019) was five times stronger than the effect of hydrological and biological processes that lead to 12 reduced bare ice albedo (Ryan et al., 2019). Impurities, in part biologically active (Ryan et al., 2018), have 13 been observed to lead to albedo reduction (Stibal et al., 2017) and are estimated to have increased runoff 14 from bare ice in the south-western sector of the Greenland Ice Sheet by about 10% (Cook et al., 2020). In 15 summary, new studies confirm that there is high confidence that the Greenland Ice Sheet melt increase since 16 about 2000 has been amplified by positive albedo feedbacks, with the expansion of bare-ice extent being the 17 dominant factor, and albedo in the bare-ice zone being primarily controlled by distributed biologically active 18 impurities (see also Section 7.3.4.3. 19 20 The SROCC reported with medium confidence that around half of the 1960-2014 Greenland Ice Sheet 21 surface meltwater ran off, while most of the remainder infiltrated firn and snow, where it either refroze or 22 accumulated in firn aquifers. Studies since SROCC show a decrease of firn air content between 1998-2008 23 and 2010-2017 (Vandecrux et al., 2019) in the low-accumulation percolation area of western Greenland, 24 reducing meltwater retention capacity. Moreover, meltwater infiltration into firn can be strongly limited by 25 low-permeability ice slabs created by refreezing of infiltrated meltwater (Machguth et al., 2016). Recent 26 observations and modelling efforts indicate that rapidly expanding low-permeability layers have led to an 27 increase in runoff area since 2001 (MacFerrin et al., 2019). In summary, there is medium confidence that 28 meltwater storage and refreezing can temporarily buffer a large-scale melt increase but limiting factors have 29 been identified. 30 31 The SROCC reported that there was medium confidence that ocean temperatures near the grounding zone of 32 tidewater glaciers are critically important to their calving rate, but there was low confidence in understanding 33 their response to ocean forcing. The increase in ice discharge in the late 1990s and early 2000s (Mouginot et 34 al., 2019; King et al., 2020; Mankoff et al., 2020) has been associated with a period of widespread tidewater 35 glacier retreat (Murray et al., 2015; Wood et al., 2021) and speed up (Moon et al., 2020). Since the SROCC, 36 new studies provide strong evidence for rapid submarine melting at tidewater glaciers (Sutherland et al., 37 2019; Wagner et al., 2019; Bunce et al., 2020; Jackson et al., 2020b). Changes in submarine melting and 38 subglacial meltwater discharge can trigger increased ice discharge by reducing the buttressing to ice flow 39 and promoting calving (Benn et al., 2017; Todd et al., 2018; Ma and Bassis, 2019; Mercenier et al., 2020); 40 through undercutting (Rignot et al., 2015; Slater et al., 2017b; Wood et al., 2018; Fried et al., 2019) and 41 frontal incision (Cowton et al., 2019). Warming ocean waters have been implicated in the recent thinning 42 and breakup of floating ice tongues in northeastern and northwestern Greenland (Mouginot et al., 2015; 43 Wilson et al., 2017; Mayer et al., 2018; Washam et al., 2018; An et al., 2021; Wood et al., 2021). On decadal 44 timescales, tidewater glacier terminus position correlates with submarine melting (Slater et al., 2019). Over 45 shorter timescales, individual glaciers or clusters of glaciers can behave differently and asynchronously 46 (Bunce et al., 2018; Vijay et al., 2019; An et al., 2021), and there are not always clear associations between 47 water temperature and glacier calving rates (Motyka et al., 2017), retreat or speedup (Joughin et al., 2020; 48 Solgaard et al., 2020). Variations in ice mélange at the front of a glacier, associated with changes in ocean 49 and air temperature, have also emerged as a plausible control on calving (Burton et al., 2018; Xie et al., 50 2019; Joughin et al., 2020). In summary, there is high confidence that warmer ocean waters and increased 51 subglacial discharge of surface melt at the margins of marine-terminating glaciers increase submarine melt, 52 which leads to increased ice discharge, and medium confidence that this contributed to the increased rate of 53 mass loss from Greenland particularly in the period 2000-2010 when increased discharge was observed in 54 the southeast and northwest. 55 Do Not Cite, Quote or Distribute 9-55 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 The SROCC reported that accurate bedrock topography is required for understanding and projecting the 2 glacier response to ocean forcing. Accurate bathymetry is essential for both establishing which water masses 3 enter glacial fjords and for reliable estimates of the submarine melt rates experienced by tidewater glaciers 4 (Schaffer et al., 2020; Slater et al., 2020b; Wood et al., 2021). Subglacial and lateral topography is known to 5 strongly modulate tidewater glacier dynamics and the sensitivity of tidewater glaciers to climatic forcing 6 (Enderlin et al., 2013; Catania et al., 2018). Bathymetric mapping around the ice sheet has greatly improved 7 with direct and gravimetric surveys (Millan et al., 2018; An et al., 2019a, 2019b; Jakobsson et al., 2020) 8 leading to the improvement of Greenland-wide bathymetric and topographic mapping (e.g., Morlighem et al., 9 2017). However, large uncertainties in ice thickness remain for around half of the outlet glaciers (Mouginot 10 et al., 2019; Wood et al., 2021) and sea-ice covered and iceberg-packed regions remain poorly sampled near 11 glacier termini (Morlighem et al., 2017). There is high confidence that bathymetry (governing the water 12 masses that flow into fjord cavities) and fjord geometry and bedrock topography (controlling ice dynamics) 13 modulate the response of individual glaciers to climate forcing. 14 15 The AR5 assessed that it is likely that anthropogenic forcing contributed to the surface melting of Greenland 16 since 1993 (Bindoff et al., 2013). Section 3.4.3.2 assesses that it is very likely that human influence has 17 contributed to the observed surface melting of the Greenland Ice Sheet over the past two decades, and there 18 is medium confidence of an anthropogenic contribution to recent mass loss from Greenland. 19 20 21 9.4.1.2 Model evaluation 22 23 The SROCC (Oppenheimer et al., 2019) stated that substantial challenges remained for modelling of both the 24 Greenland surface mass balance and the dynamical ice sheet. Since the SROCC, further insights into 25 modelling of the Greenland ice sheet has come from model intercomparison studies of both the surface mass 26 balance (Fettweis et al., 2020) and dynamical ice sheets (Goelzer et al., 2020). Further aspects relevant to the 27 forcing of the ice sheet from large scale global climate models and regional climate models are discussed in 28 Box 9.3 and Section Atlas.11.2. 29 30 The SROCC stated that climate model simulations of Greenland surface mass balance (SMB) had improved 31 since the AR5 giving medium confidence in the ability of climate models to simulate changes in Greenland 32 SMB. Since the SROCC, a multi-model intercomparison study (Fettweis et al., 2020) of regional and global 33 climate models has shown that the greatest inter-model spread occurs in the ablation zone, due to 34 deficiencies in an accurate model representation of the ablation zone extent and processes related to surface 35 melt and runoff, confirming the SROCC statement that bare-ice model uncertainty is large (Ryan et al., 36 2019). This intercomparison showed that simple, well-tuned SMB models using positive degree-day melt 37 schemes can perform as well as more complex physically-based models (Figure Atlas.30). Furthermore, the 38 ensemble-mean of the models produced the best estimate of the present-day SMB relative to observations 39 (particularly in the ablation zone). Further assessment of Greenland ice sheet regional SMB can be found in 40 Section Atlas 11.2.3. Recent progress confirms the SROCC assessment that there is medium confidence in 41 the ability of climate models to simulate changes in Greenland SMB. 42 43 The SROCC noted increased use of coupled climate-ice sheet models for simulating the Greenland ice sheet, 44 but also that remaining deficiencies in coupling between models of climate and ice sheets (e.g., low spatial 45 resolution) limited the adequate representation of the feedbacks between them. Some Earth System Models 46 now incorporate multi-layer snow models and full energy balance models (Punge et al., 2012; Cullather et 47 al., 2014; van Kampenhout et al., 2017; Alexander et al., 2019a; van Kampenhout et al., 2020) or use 48 elevation classes to compensate for their coarser resolution (Lipscomb et al., 2013; Sellevold et al., 2019; 49 Muntjewerf et al., 2020a, 2020b; Gregory et al., 2020). Resulting SMB simulations compare better with 50 Regional Climate Models and observations (Alexander et al., 2019b; van Kampenhout et al., 2020), but 51 remaining shortcomings lead to problems reproducing a present-day ice sheet state close to observations. In 52 summary, there is medium confidence in quantitative simulations of the present-day state of the Greenland 53 ice sheet in ESMs. 54 55 The SROCC (Meredith et al., 2019) stated that there is low confidence in understanding coastal glacier Do Not Cite, Quote or Distribute 9-56 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 response to ocean forcing because submarine melt rates, calving rates, bed and fjord geometry and the roles 2 of ice mélange and subglacial discharge are poorly understood. Ice-ocean interactions remain poorly 3 understood and difficult to model, with parameterizations often used for calving of marine-terminating 4 glaciers (Mercenier et al., 2018) and submarine and plume-driven melt (Beckmann et al., 2019). Due to the 5 difficulties of modelling the large number of marine-terminating glaciers and limited availability of high- 6 resolution bedrock data, the majority of recent modelling work on Greenland outlet glaciers is focused on 7 individual or a limited number of glaciers (Krug et al., 2014; Bondzio et al., 2016; Morlighem et al., 2016a; 8 Muresan et al., 2016; Bondzio et al., 2017; Choi et al., 2017; Beckmann et al., 2019), or a specific region 9 (Morlighem et al., 2019). Since the SROCC, using a flowline model that includes calving and submarine 10 melting, Beckmann et al. (2019) concluded that the AR5 upscaling of contributions from four of the largest 11 glaciers (Nick et al., 2013) overestimated the total glacier contribution from the Greenland ice sheet, due to 12 differences in response between large and small glaciers. The regional study of Morlighem et al. (2019) 13 confirms that ice-ocean interactions have the potential to trigger extensive glacier retreat over decadal 14 timescales as indicated by observations (Section 9.4.1.1). A focus in continental ice sheet models has been 15 the improved treatment of marine-terminating glaciers via inclusion of calving processes and freely moving 16 calving fronts (Aschwanden et al., 2019; Choi et al., 2021). An improved bedrock topographic dataset 17 (Morlighem et al., 2017) allows for ice discharge to be better captured for outlet glaciers in continental ice 18 sheet models and simulations indicate that bedrock topography controls the magnitude and rate of retreat 19 (Aschwanden et al., 2019; Rückamp et al., 2020). Overall, although there is high confidence that the dynamic 20 response of Greenland outlet glaciers is controlled by bedrock topography, there is low confidence in 21 quantification of future mass loss from Greenland triggered by warming ocean conditions due to limitations 22 in current understanding of ice-ocean interactions, its implementation in ice sheet models, and knowledge of 23 bedrock topography. 24 25 The SROCC (Oppenheimer et al., 2019) noted the progress made in Greenland Ice Sheet models since the 26 AR5. New since the SROCC is a focus on improved representation of the present-day state of the ice sheet 27 (Goelzer et al., 2018, 2020) (Box 9.3). Improvements are closely linked to the growing number and quality 28 of observations (Section 9.4.1.1), new techniques to generate internally consistent input data sets 29 (Morlighem et al., 2014, 2016b), wider use of data assimilation techniques (Larour et al., 2014, 2016; Perego 30 et al., 2014; Goldberg et al., 2015; Lee et al., 2015; Schlegel et al., 2015; Mosbeux et al., 2016), increased 31 model resolution (Aschwanden et al., 2016) and tuning of key processes such as calving (Choi et al., 2021). 32 A remaining challenge is low confidence in reproducing historical mass changes of the Greenland Ice Sheet 33 (Box 9.3). However, there is medium confidence in ice sheet models reproducing the present state of the 34 Greenland Ice Sheet leading to medium confidence in current ability to accurately project its future evolution. 35 36 37 9.4.1.3 Projections to 2100 38 39 The AR5 and the SROCC projected that changes in Greenland surface mass balance (SMB) will contribute 40 to sea level in 2100 by 0.03 (0.01 to 0.07) m SLE under RCP2.6 and 0.07 (0.03 to 0.16) m SLE under 41 RCP8.5. New since the SROCC are projections of SMB obtained by an Earth system model, two regional 42 climate models, and reconstructions based on temperature from the CMIP5 and CMIP6 ensembles (Hofer et 43 al., 2020; Noël et al., 2021). The range of sea level contribution from Greenland surface mass balance in 44 Noël et al. (2021) are comparable to the AR5 assessment when either CMIP5 or CMIP6 models are used 45 while Hofer et al., (2020) find a greater mass loss across all CMIP6 emission scenarios when compared to 46 CMIP5 scenarios. Using SSP5-8.5 instead of RCP8.5 increases the mean projected sea level from 2005-2100 47 by up to 0.06 m in the regional climate model simulations of Hofer et al., (2020) who attribute the difference 48 mainly to a greater Arctic amplification and associated cloud and sea ice feedbacks in the CMIP6 SSP5-8.5 49 simulations. In summary, these new projections with fixed ice sheet topography do not provide sufficient 50 evidence to change the AR5 and SROCC assessment. 51 52 Reviewing modelling studies since the AR5 (Church et al., 2013a), the SROCC (Oppenheimer et al., 2019) 53 assessed Greenland’s contribution to future sea level to be relatively similar to the AR5 (Table 9.2). The 54 baseline for projections has shifted from 1986-2005, in the SROCC, to 1995-2014 in this report. Adjusted to 55 the new 1995-2014 baseline by subtracting 0.01 m, the SROCC projected a likely contribution of 0.07 (0.03 Do Not Cite, Quote or Distribute 9-57 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 – 0.11) m SLE under RCP2.6 and 0.14 (0.08 – 0.27) m SLE under RCP8.5 by 2100. Since the SROCC, new 2 projections for the 21st century have included dynamic ice sheets coupled to Earth system models 3 (Muntjewerf et al., 2020a; Van Breedam et al., 2020) or regional atmospheric models (Le clec’h et al., 2019) 4 (Table 9.2). The coupled ESM-ice sheet CESM2-CISM2 model projects a sea level rise of 0.109 m in 2100 5 relative to 2015 under SSP5-8.5 (Muntjewerf et al., 2020a) and a similar contribution under the idealized 1% 6 yr-1 increase in CO2 scenario (Muntjewerf et al., 2020b). The CESM2-CISM2 simulations include ice sheet- 7 atmosphere interactions and ice sheet surface meltwater is routed to the ocean. The coupled regional 8 atmospheric model and ice sheet model (MAR-GRISLI) projects a sea level rise of 0.079 m in 2100 relative 9 to 2000 under RCP8.5 (Le clec’h et al., 2019). An Earth System model of lower complexity coupled to an 10 ice sheet model gives a sea level contribution of 0.025 to 0.064 m under RCP2.6 and 0.056 to 0.12 m under 11 RCP8.5 (the range is due to four simulations with different parameter sets for the atmosphere model) (Van 12 Breedam et al., 2020). Van Breedam et al., (2020) identify a simulation with a preferred parameter set, that 13 projects 0.034 m for RCP2.6and 0.073 m for RCP8.5 Although the ocean does not directly force the ice sheet 14 models in these simulations, the new coupled models allow for interactions between ice sheet dynamics, 15 surface mass balance and local climate. The coupled projections fall within the lower bounds of the AR5 and 16 the SROCC, and as these studies do not prescribe ocean forcing directly, it is possible that the dynamic 17 response is underestimated. 18 19 Since the SROCC, projections of the Greenland Ice Sheet are also now available from ISMIP6 (Nowicki et 20 al., 2016, 2020a; Box 9.3; Annex II; Figure 9.17). The ISMIP6 multi-model projections are corrected with an 21 assessment of the historical dynamical response to pre-2015 climate forcing (Box 9.3). For the period 2015- 22 2100, the ISMIP6 uncorrected multi-model ensemble projects a sea level contributions ranging from 0.01 to 23 0.05 m under RCP2.6, 0.04 to 0.14 m under RCP8.5 (Goelzer et al., 2020), 0.02 to 0.06 m under SSP1-2.6 24 and 0.08 to 0.25 m under SSP5-8.5 (Payne et al., 2021)(Table 9.2). The higher mass loss in the SSPs is 25 attributed to a larger decrease in SMB due to the high climate sensitivity of the models used (Payne et al., 26 2021). This finding is confirmed by (Choi et al., 2021), where CMIP6 SSP5-8.5 SMB leads to larger ice loss 27 than CMIP5 RCP8.5 while ice discharge is similar. As the ISMIP6 framework considers a subset of the 28 RCPs/SSPs and CMIP models, SSP-based projections have been inferred from multiple approaches. First, 29 the ISMIP6 CMIP5-forced (Goelzer et al., 2020) and CMIP6-forced (Payne et al., 2021) combined ensemble 30 projections were corrected with the historical trend (Box 9.3) using bootstrapping. Second, an emulator of 31 the ISMIP6 projections (Edwards et al., 9998, Box 9.3) is forced by distributions of global surface air 32 temperature for each SSP from a two-layer energy budget emulator (Supplementary Material 7.SM.2) and 33 then corrected with the historical trend in the same way. These two approaches result in projections that are 34 similar in their median values to the AR5 and SROCC projections (Table 9.2), but differ in their range. 35 Similar results are obtained when the AR5 parametric fit is applied to the ISMIP6 models (Table 9.2, 36 Supplementary Material 9.SM.4.4), which is used to estimate rates of change and post-2100 projections 37 (Sections 9.4.1.4, 9.6.3.2). 38 39 The SROCC noted that the study by Aschwanden et al.,( 2019) projects a significantly higher Greenland 40 contribution to sea level than the assessed likely range in the AR5 and the SROCC. Under RCP8.5, 41 Aschwanden et al., (2019) found that Greenland could contribute up to 0.33 m to sea level by 2100 relative 42 to 2000 (the ensemble member that best reproduces 2000-2015 mean surface mass balance from a regional 43 climate model projects Greenland mass losses of 0.08 m SLE under RCP2.6 and 0.18 m SLE under RCP8.5.) 44 The SROCC noted that the potentially high sea level in this study could be due to the assumption of spatially 45 uniform warming which can overestimate surface melt rates. However, it also reflects the deep uncertainty 46 surrounding atmospheric forcing, surface processes, submarine melt, calving and ice dynamics. Goelzer et 47 al., (2020) ascribe 40% of the ISMIP6 multi-model ensemble spread to ice sheet model uncertainty, 40% to 48 climate model uncertainty and 20% to ocean forcing uncertainty. We note that this finding reflects the 49 current challenges associated with the representation of ice-ocean interactions in models, and the uncertainty 50 in basal conditions (Section 9.4.1.2). However, this finding is consistent with the work of Aschwanden et al., 51 (2019) and thus, there is medium confidence that uncertainty in mass loss from the Greenland Ice Sheet is 52 dominated by uncertainty in climate scenario and surface processes, whereas uncertainty in calving and 53 frontal melt play a minor role. 54 55 The SROCC stated that surface processes, rather than ice discharged into the ocean, will dominate Greenland Do Not Cite, Quote or Distribute 9-58 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 ice loss over the 21st century, regardless of the emissions scenario (high confidence). This is confirmed by 2 the ISMIP6 projections (Goelzer et al., 2020; Payne et al., 2021). The projected mass loss of Greenland is 3 predominantly due to increased surface meltwater and loss in refreezing capacity resulting in decreasing 4 SMB (high confidence), concurrent with rising temperatures and darkening of the ice sheet surface (Fettweis 5 et al., 2013; Vizcaino et al., 2015; Le clec’h et al., 2019; Muntjewerf et al., 2020b, 2020a; Sellevold and 6 Vizcaíno, 2020). Mass changes due to SMB and outlet glacier dynamics are linked (Goelzer et al., 2013; 7 Fürst et al., 2015; Rückamp et al., 2020) as mass loss by one process decreases mass loss by the other (for 8 example, SMB removes ice before it can reach the marine glacier terminus). There is medium confidence that 9 the mass loss through ice discharge will decrease in the future (Fürst et al., 2015; Aschwanden et al., 2019; 10 Golledge et al., 2019a), because an increase in mass loss (via increased discharge or surface runoff) leads, in 11 most areas, to a retreat of the glacier margin onto land above sea level, isolating the ice sheet from marine 12 influence. 13 14 In summary, it is virtually certain that the Greenland Ice Sheet will continue to lose mass this century under 15 all emissions scenarios and high confidence that total mass loss by 2100 will increase with cumulative 16 emissions. The sea level assessment (Section 9.6.3.3) is based on the emulated ISMIP6 projections allowing 17 a more consistent approach to a wider range of climate and ocean forcings. The Greenland ice sheet is likely 18 to contribute 0.06 (0.01 to 0.10) m under SSP1-2.6 and 0.13 (0.09 to 0.18) m under SSP5-8.5 by 2100 19 relative to 1995-2014. These projections (as well as those of AR5 and SROCC) are lower than the study of 20 (Aschwanden et al., 2019) (or the range of possible sea level changes resulting from structured expert 21 judgement (Bamber et al., 2019); Section 9.6.3.2), contributing to the deep uncertainty in projected sea level 22 (Box 9.4). There is however high confidence that the loss from Greenland will become increasingly 23 dominated by SMB and surface melt, as the ocean-forced dynamic response of glaciers will diminish as 24 marine margins retreat to higher grounds. 25 26 27 [START TABLE 9.2 HERE] 28 29 Table 9.2: Projected sea level contributions in meters from the Greenland ice sheet by 2100 relative to 1995-2014, 30 unless otherwise stated, for selected RCP and SSP scenarios. Italics denote partial contributions. 31 Historical dynamic response omitted from ISMIP6 simulations is estimated to be 0.19 ± 0.10 mm yr-1 32 (0.02 m ± 0.01 m in 2100 relative to 2015). The climate forcing is described in Appendix 7.SM.2. 33 Representative Concentration Pathways (RCPs) Study RCP2.6 RCP4.5 RCP8.5 Notes IPCC AR5 and 0.07 (0.03 to 0.08 (0.04 to 0.14 (0.08 to Median and likely SROCC 0.11) 0.15) 0.27) (66% range) (Oppenheimer contributions in et al., 2019) 2100 relative to 1995-2014. Median of multiple studies. ISMIP6 0.01 to 0.05 --- 0.04 to 0.14 Range of multi- CMIP5-forced model (Goelzer et al., contributions in 2020); excludes 2100 relative to historical 2015 from 1 ESM dynamic for RCP2.6 and 6 response ESMs for RCP8.5. (see caption). Coupled --- --- 0.079 Contribution in regional 2100 relative to atmosphere-ice 2000 from MAR- sheet model (Le GRISLI model. clec’h et al., Do Not Cite, Quote or Distribute 9-59 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 2019) Coupled Earth 0.034 (0.025 to 0.073 (0.056 to Contribution in system model of 0.064) 0.12) 2100 relative to lower 2000 from complexity-ice LOVECLIM- sheet model AGISM model. (Van Breedam Preferred et al., 2020) parameter set and range from 4 simulations with different parameters for atmosphere model. Shared Socioeconomic Pathways (SSPs) Study SSP1-2.6 SSP2-4.5 SSP5-8.5 Notes Coupled Earth --- --- 0.109 Contribution in system model- 2100 relative to ice sheet model 2015 from coupled (Muntjewerf et CESM2-CISM2. al., 2020, a) ISMIP6 0.02 to 0.06 --- 0.08 to 0.25 Range of multi- CMIP6-forced model (Payne et al., contributions in 2021); excludes 2100 relative to historical 2015 from 1 ESM dynamic for SSP1-2.6 and 4 response ESMs for SSP5- 8.5. ISMIP6 CMIP5 0.06 (0.05 to --- 0.11 (0.09 to Median (66% and CMIP6 0.07) 0.14) range) [90% range] forced ensemble [0.04 to 0.08] [0.07 to 0.17] contribution from including ISMIP6 CMIP5- historical and CMIP6-forced dynamic multi-model response ensembles ISMIP6 with 0.08 (0.06 to 0.10 (0.08 to 0.14 (0.11 to Median (66% AR5 parametric 0.10) 0.13) [0.07 to 0.18) range) [90% range] fit: used to [0.05 to 0.12] 0.15] [0.10 to 0.22] contribution from estimate rates AR5 parametric fit (Supplementary to ISMIP6 Material ensemble, relative 9.SM.4.4) to 1995-2014. including historical dynamic response. Emulated 0.03 (-0.01 to 0.06 (0.01 to 0.11 (0.06 to Median (66% ISMIP6; 0.08) 0.10) 0.16) range) [90% excludes [-0.04 to 0.12] [-0.02 to 0.15] [0.03 to 0.21] range] contribution historical in 2100 relative to dynamic 2015 from response emulator of (Edwards et al., ISMIP6 used with Do Not Cite, Quote or Distribute 9-60 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 2021) chapter 7 climate forcing. This 0.06 (0.01 to 0.08 (0.04 to 0.13 (0.09 to As above, but assessment: 0.10) 0.13) 0.18) relative to 1995- emulated [-0.02 to 0.15] [0.01 to 0.18] [0.05 to 0.23] 2014 and ISMIP6 total including historical dynamic response. 1 2 [END TABLE 9.2 HERE] 3 4 5 9.4.1.4 Projections beyond 2100 6 7 The AR5 (Church et al., 2013a) assessed the contribution from Greenland to sea level projections in 2300 as 8 0.15 m SLE in low emissions scenarios (~RCP2.6) and 0.31-1.19 m in high scenarios (~RCP6.0/RCP8.5). 9 The SROCC (Oppenheimer et al., 2019) did not update the AR5 estimates given limited evidence and low 10 agreement from three new studies (Vizcaino et al., 2015; Calov et al., 2018; Aschwanden et al., 2019). Since 11 the SROCC, a new study gives a sea level contribution of 0.11 to 0.20 m in low- and 0.61 to 1.29 m in high- 12 emissions scenarios (Van Breedam et al., 2020). The low emissions projections by Van Breedam et al., 13 (2020) encompass the AR5 assessed contribution, while the high emissions are higher than that from the 14 AR5. The ‘optimal’ ensemble member of Aschwanden et al., (2019) (Section 9.4.1.3) indicates that 15 Greenland could contribute 0.25 m under RCP2.6 and 1.74 m under RCP8.5. Structured expert judgement 16 (Bamber et al., 2019) projects Greenland losses of 0.54 (0.28-1.28) m under 2ºC warming and 0.97 (0.4- 17 2.23) m under 5ºC warming. These studies therefore agree that the AR5 and SROCC assessments are at the 18 low end of the range of projections. In addition, observations suggest that Greenland Ice Sheet losses are 19 tracking the upper range of AR5 projections (Slater et al., 2020b). Therefore, we update the likely range for 20 the contribution of the Greenland Ice Sheet to GMSL by 2300 to 0.11-0.25 m under RCP2.6 and 0.31-1.74 m 21 under RCP8.5. However, given the uncertainty in climatic drivers used to project ice sheet change over the 22 21st century (Goelzer et al., 2020; Hofer et al., 2020; Noël et al., 2021) and the large range in simulations 23 since AR5 extending beyond 2100, we only have low confidence in the contribution to GMSL by 2300 and 24 beyond. 25 26 27 The role of the elevation-mass feedback for future projections of Greenland can be assessed from paleo 28 simulations. Ice sheet model simulations of the Laurentide (Gomez et al., 2015; Gregoire et al., 2016) and 29 Eurasian (Alvarez-Solas et al., 2019) ice sheets invoke at least some contribution to last glacial termination 30 mass loss from SMB reduction, as a consequence of an elevation–mass balance feedback (Levermann and 31 Winkelmann, 2016). In a model spanning Meltwater Pulse 1A this mechanism increased mass loss by 32 approximately 66% (Gregoire et al., 2016) but in Last Interglacial simulations the effect of this feedback is 33 shown to depend on the surface scheme of the climate model employed (Plach et al., 2019). Given the 34 agreement between theoretical analyses and paleo-ice sheet model experiments there is high confidence that 35 the elevation-mass balance feedback is most relevant at multi-centennial and millennial timescales, 36 consistent with future-focused studies (Gregory et al., 2020, Aschwanden et al. 2019, Le clec’h et al., 2019). 37 38 The SROCC adopted the AR5 assessment that complete loss of Greenland ice, contributing about 7 m to sea 39 level, over a millennium or more would occur for a sustained GMST between 1ºC (low confidence) and 4ºC 40 (medium confidence) above pre-industrial levels. New studies since the SROCC (Gregory et al., 2020; Van 41 Breedam et al., 2020) confirm this assessment (see also Figure 9.30). Clark et al., (2016) estimate a complete 42 loss to take about 8000 years at 5.5°C and about 3000 years at 8.6°C. Based on agreement between new and 43 previous studies there is therefore high confidence that the rate at which Greenland ice sheet commitment is 44 realized depends upon the amount of warming. 45 46 Accounting for more detailed feedbacks between the atmosphere and the ice sheet (Gregory et al., 2020) 47 found a gradual relationship between sustained global mean warming and the corresponding near- Do Not Cite, Quote or Distribute 9-61 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 equilibrium ice sheet volume, in contrast to a sharp threshold as found by (Robinson et al., 2012). Rather 2 than a climatically-controlled tipping point for irreversible loss of the Greenland Ice Sheet, (Gregory et al., 3 2020) found a threshold of irreversibility linked to ice sheet size, similar to previous work (Ridley et al., 4 2010). The results of Gregory et al., (2020) show that if the ice sheet loses mass equivalent to about 3-3.5 m 5 of sea-level rise, it would not regrow to its present state, and 2m of the sea-level rise would be irreversible. 6 At which point in time the current ice sheet might reach this critical volume depends on oceanic and 7 atmospheric conditions, ice dynamics, and climate-ice sheet feedbacks (Gregory et al., 2020; Van Breedam 8 et al., 2020), so projections differ in the magnitude and rate of temperature change to cross the threshold for 9 irreversible loss. Projections from a large ensemble indicate that the mass threshold may be reached as early 10 as in 400 years, if warming reaches as high as >10°C above present under extended RCP8.5 (Aschwanden et 11 al., 2019). In summary, there is high confidence in the existence of threshold behaviour of the Greenland Ice 12 Sheet in a warmer climate, however there is low agreement on the nature of the thresholds and the associated 13 tipping points. 14 15 16 [START BOX 9.3 HERE] 17 18 BOX 9.3: Insights into land ice evolution from model intercomparison projects 19 20 Projections of ice sheets and glaciers in the AR5 (Church et al., 2013a) and the SROCC (Oppenheimer et al., 21 2019) were assessed by collecting single model studies (with the exception of glaciers in SROCC (Hock et 22 al., 2019b)). Community benchmark experiments (ISMIP-HOM; (Pattyn et al., 2008) or Marine Ice Sheet 23 Model Intercomparison Projects (MISMIP, (Pattyn et al., 2012); MISMIP3d, (Pattyn and Durand, 2013); 24 MISMIP+ (Asay-Davis et al., 2016; Cornford et al., 2020)) have substantially advanced ice sheet modelling 25 since the AR5. Model Intercomparison Projects (MIPs) now inform projections of both ice sheets and 26 glaciers: the Ice Sheet MIP for CMIP6 (ISMIP6) (Section 9.4.1.3; 9.4.2.5), the Linear Antarctic Response 27 MIP (LARMIP-2; Section 9.4.2.5) and GlacierMIP (Section 9.5.1.3). 28 29 Regional forcing for land ice intercomparison projects 30 Simulations of ice sheets and glaciers are dependent on forcing provided by atmosphere and ocean 31 models. Despite progress in representing processes, reducing biases and increasing resolution, regional and 32 global models still have difficulties reproducing observed regional air temperature, surface mass balance and 33 ocean changes (Section 9.4.1.2, 9.4.2.2, Atlas.11). Assessment of CMIP5 and CMIP6 climate models, as 34 forcing for land ice models, has been undertaken (Walsh et al., 2018; Barthel et al., 2020a; Marzeion et al., 35 2020; Nowicki et al., 2020b) with the aim of selecting the best available historical forcings and sampling 36 potential regional future climate changes. Despite improvement in simulation of atmospheric forcing, 37 persistent biases remain in CMIP5/6, which reduces the fidelity of historical and future simulations of land 38 ice. 39 40 ISMIP6 initial state intercomparison projects (initMIP) 41 The ISMIP6 initial state intercomparison projects (initMIP) for the Greenland (Goelzer et al., 2018) and 42 Antarctic (Seroussi et al., 2019) ice sheets were designed to understand the uncertainty in sea level 43 projections resulting from the choice of initialization procedures used for projections of sea level (Nowicki et 44 al., 2016). Participating modelling groups (Annex II) were free to decide on the initialization method used to 45 bring ice sheet models to a present-day state with the effect of these choices captured in a control simulation 46 (starting from the present day state, with no further climate forcing applied), which measures intrinsic model 47 drift. Compared to the earlier SeaRISE intercomparison project (Bindschadler et al., 2013; Nowicki et al., 48 2013), the modelled present day ice sheets are in closer agreement with observations and the model drift has 49 been reduced (Goelzer et al., 2018; Seroussi et al., 2019). Nonetheless, historical simulations remain 50 challenging for ice sheet models, due to limited ice sheet observations prior to the satellite era and biases in 51 the historical atmospheric and oceanic forcings from climate models (Nowicki and Seroussi, 2018).ISMIP6 52 and LARMIP-2 therefore did not provide a protocol for the historical runs used to bring the ice sheets to 53 present-day, nor criteria for sub-selecting models from the multi-model ensemble based on ability to 54 reproduce historical changes (Levermann et al., 2020a; Nowicki et al., 2020a). 55 Do Not Cite, Quote or Distribute 9-62 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 ISMIP6 projections for the Greenland and Antarctic ice sheets 2 The ISMIP6 projection protocol (Nowicki et al., 2016, 2020a) was designed to sample the uncertainty in 3 future sea level due to climate scenarios (via the use of high and low emission scenarios and multiple climate 4 models), ice-ocean interactions and inland response to ice shelf collapse, and ice sheet model diversity. The 5 participanting ice sheet models are listed in Annex II. For each ice sheet, forcing was selected (Barthel et al., 6 2020b) from the CMIP5 (Taylor et al., 2012) and CMIP6 (Eyring et al., 2016) models. Atmospheric forcing 7 fields consisted of anomalies in surface mass balance and surface air temperatures generated directly from 8 the CMIP models for the Antarctic ice sheet and downscaled using the MAR regional climate model for the 9 Greenland ice sheet (Hofer et al., 2020). To sample the uncertainty due to ocean forcings, models used either 10 a model-specific scheme using the ISMIP6-provided oceanic dataset or a standard ISMIP6 approach. For the 11 Greenland ice sheet, the oceanic dataset consists of thermal forcing (temperature minus freezing 12 temperature) extrapolated into fjords and subglacial runoff. The standard approach uses timelines of 13 tidewater glacier retreat (Slater et al., 2019, 2020a). For the Antarctic ice sheet, the oceanic dataset consists 14 of salinity, thermal forcing and temperature added to an observationally derived climatology and 15 extrapolated under ice shelves. The standard approach is a basal melt rate that depends quadratically on 16 thermal forcing, adapted from Favier et al., (2019), with two different calibrations (Figure 9.19, Jourdain et 17 al., 2020) that reproduce observed basal melt rates across Antarctica or Pine Island Glacier, respectively 18 (Sections 9.4.2.2, 9.4.2.3). Antarctic ice shelf disintegration datasets (Nowicki et al., 2020a) assume that ice 19 shelves disintegrate when annual surface melt reaches a threshold (Trusel et al., 2015). 20 21 The ISMIP6 projections (Goelzer et al., 2020; Seroussi et al., 2020; Payne et al., 2021) are reported as 22 experiment minus control and represent the sea level resulting from future climate change only. The control 23 simulation, which has constant climate conditions starting in 2015 from the historical run, captures drift 24 associated with the choices made for the initialization method and historical run. Subtraction of this control 25 removes any long-term dynamic response of the ice sheet to pre-2015 climate change. This response has 26 been assessed using dynamic discharge derived from observations over the last 40 years (Mouginot et al., 27 2019; Rignot et al., 2019), under an assumption that it persists at the past rate until 2100 rather than 28 diminishing. The dynamic response to historical forcing is estimated as 0.19 ± 0.10 mm yr-1 for the 29 Greenland ice sheet (Section 9.4.1.3) and 0.33 ± 0.16 mm yr-1 for the Antarctic ice sheet (Section 9.4.2.5). 30 Over the period 2015-2100, this leads to an additional sea level contribution of 1.7 cm for Greenland and 2.8 31 cm for Antarctica. 32 33 LARMIP-2 projections for the Antarctic ice sheet 34 LARMIP-2 is focused on the uncertainty in the ocean forcing and associated ice shelf melting (Levermann et 35 al., 2014, 2020a) with the majority of the models also participating in ISMIP6 (Annex II). The experiments 36 start from present day and impose an additional basal ice shelf melting of 8 m yr-1 at the beginning of the 100 37 yr simulation. A control run is used to remove drift resulting from initialization. The time derivative of the 38 ice sheet response yields a linear response function, which is then convoluted with a forcing of basal shelf 39 melt time series for five Antarctic regions. The forcing time series for RCP2.6, 4.5, 6.0 and 8.5 were 40 obtained from a random combination of global mean temperature for each RCP from MAGICC-6.0 41 (Meinshausen et al., 2011), a scaling factor and time delay for the relationship between global surface air 42 temperature and subsurface ocean warming in a given sector of the Southern Ocean from one of 19 CMIP5 43 models (Taylor et al., 2012) and a basal melting sensitivity from the interval [7,- 16] m yr-1 ºC-1 to convert 44 the regional subsurface warming into basal ice shelf melting. This process is repeated 20,000 times to obtain 45 a probability distribution of the sea level contribution for five Antarctic sectors. The linear response 46 framework captures complex temporal responses of the ice sheets resulting from an increase in basal ice 47 shelf melting, but neglects the response to surface mass balance and any self-dampening or self-amplifying 48 processes, such as MISI. The LARMIP-2 method is applied to temperature projections for the SSPs 49 (Supplementary Material 7.SM.2) and an estimate of surface mass balance change from the AR5 parametric 50 Antarctic ice sheet surface mass balance model (Church et al., 2013a) is added to the results (Sections 51 9.4.2.4, 9.4.2.5, 9.6.3.2). It is not necessary to add a long-term dynamic response to the LARMIP-2 52 projections, as this is incorporated in the basal melt time series. 53 54 GlacierMIP projections 55 GlacierMIP (Marzeion et al., 2020) was designed to estimate the glacier contribution to sea level rise, Do Not Cite, Quote or Distribute 9-63 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 including from peripheral glaciers in Greenland and Antarctica that can be considered to be dynamically 2 decoupled, or entirely separate, from the ice sheets. Glacier models are described in Annex II. Initial 3 conditions were based on version 6 of the Randolph Glacier Inventory (RGI Consortium, 2017) and initial 4 ice thickness and volume were provided from an update of Huss and Farinotti, (2012) (although some glacier 5 models used their own estimates). Forcings were taken from ten different CMIP5 GCMs, selected based on 6 availability of multiple RCPs, the choice in a previous model intercomparison (Hock et al., 2019a), and 7 performance in glacier-covered regions according to Walsh et al., (2018). In addition, two global glacier 8 models performed the same experiment with thirteen CMIP6 models (Section 9.5.1.3). 9 10 Use of an emulator with ISMIP6 and GlacierMIP projections 11 The ISMIP6 and GlacierMIP projections are primarily based on a limited number of CMIP5 RCPs and 12 CMIP6 SSPs, and a limited sampling of ice-ocean interaction parameters and ice shelf collapse simulations. 13 Emulators provide a method for expanding these projections to a range of SSPs with more comprehensive 14 sampling of climate, ice sheet and glacier modelling uncertainties. Sections 9.4.1.3, 9.4.2.5 and 9.5.1.3 show 15 estimates from the emulator of (Edwards et al., 2021). This is a Gaussian Process (rather than physically- 16 based: Cross-Chapter Box 7.1) model derived from the ISMIP6 and GlacierMIP simulations; projections use 17 distributions of GSAT from the two-layer emulator (Supplementary Material 7.SM.2) and ice sheet 18 parameters as inputs, and include estimates of the emulator uncertainty. Probability intervals are therefore 19 not inflated by a further factor, as is often the case for multi-model ensemble projections, to account for 20 missing uncertainties (Section 9.6.3.2). The emulator is used in Section 9.6.3 to provide projections of the 21 land-ice contribution to sea level that are fully consistent with each other, ocean heat content, and the 22 assessed equilibrium climate sensitivity and projections of GSAT across the entire report. 23 24 [END BOX 9.3 HERE] 25 26 27 9.4.2 Antarctic Ice Sheet 28 29 9.4.2.1 Recent observed changes 30 31 As stated in Section 2.3.2.4, satellite observations by IMBIE combining multi-team estimates based on 32 altimetry, gravity anomalies (GRACE) and the input-output method, already presented in the SROCC 33 (Meredith et al., 2019), is updated and extended to 2020 (The IMBIE team, 2021). The Antarctic ice sheet 34 (AIS) lost 2670 [1800–3540] Gt mass over the period 1992–2020, equivalent to 7.4 [5.0–9.8] mm global 35 mean sea level rise (see Table 9.5 for contribution to sea-level budget and Figures 9.16, 9.18). Within 36 uncertainties, this estimate agrees with a review of post-AR5 studies up to 2016 (Bamber et al., 2018b) and 37 is consistent with recent single studies based on satellite laser altimetry (Smith et al., 2020), the input-output 38 method (Rignot et al., 2019) and gravimetry (Velicogna et al., 2020). The mass-loss rate was on average 49 39 [-2 to 100] Gt yr-1 over the period 1992–1999, 70 [22 to 119] Gt yr-1 over the period 2000–2009 and 148 [94 40 to 202] Gt yr-1 over the period 2010–2016 (see Figures 9.16, 9.18 and Table 9.SM.1). However, recent work 41 suggests that the mass loss has not further increased since 2016 because of regional mass gains in Dronning 42 Maud Land (Velicogna et al., 2020). Mass loss of the West Antarctic and Antarctic Peninsula Ice Sheets has 43 increased since about 2000 (very high confidence), essentially due to increased ice discharge (Harig and 44 Simons, 2015; Paolo et al., 2015; Forsberg et al., 2017; Bamber et al., 2018b; Gardner et al., 2018; The 45 IMBIE Team et al., 2018; Rignot et al., 2019) 46 47 The SROCC reported with very high confidence that the acceleration, retreat and thinning of the principal 48 West Antarctic outlet glaciers has dominated the observed Antarctic mass loss over the last decades, and 49 stated with high confidence that these losses were driven by melting of ice shelves by warm ocean waters. 50 The average West Antarctic Ice Sheet (WAIS) mass loss of 82 ± 9 Gt yr-1 between 1992 and 2017 (The 51 IMBIE Team et al., 2021) leads to substantial observed surface lowering (e.g., Schröder et al., 2019; 52 Shepherd et al., 2019), particularly in coastal regions (Figure 9.18). Recent studies using satellite altimetry 53 (Schröder et al., 2019) and the input-output method (Rignot et al., 2019) consistently show mass loss in these 54 coastal regions since the late 1970s (Figure 9.16). Because of consistent multiple lines of evidence, there is 55 high confidence in mass loss of the Totten Glacier in East Antarctica (Miles et al., 2013; Li et al., 2016b; Do Not Cite, Quote or Distribute 9-64 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Mohajerani et al., 2018; Rignot et al., 2019; Schröder et al., 2019; Shepherd et al., 2019) since about 2000, 2 dominated by changes in coastal ice dynamics (Li et al., 2016b). It is currently unclear whether mass loss of 3 the East Antarctic Ice Sheet (EAIS) over the last three decades has been significant (Rignot et al., 2019) or, 4 at 5 ± 46 Gt yr-1 between 1992 and 2017, essentially zero within uncertainties (The IMBIE Team et al., 5 2018). In summary, WAIS losses, through acceleration, retreat and thinning of the principal outlet glaciers, 6 dominated the AIS mass losses over the last decades (very high confidence) and there is high confidence that 7 this is the case since the late 1970s. Furthermore, parts of the EAIS have lost mass in the last two decades 8 (high confidence). 9 10 As stated in the SROCC, snowfall and glacier flow are the largest components determining AIS mass 11 changes, with glacier flow acceleration (dynamic thinning) on the WAIS and the Antarctic Peninsula driving 12 total loss trends in recent decades (very high confidence), and a partial offset of the dominating dynamic- 13 thinning losses by increased snowfall (high confidence). The SROCC attributed medium confidence to 14 estimates of 20th-century snowfall increases equivalent to a sea level change of -7.7 ± 4.0 mm on the EAIS 15 and -2.8 ± 1.7 mm on the WAIS, respectively (Medley and Thomas, 2019). Loss of buttressing, which can be 16 caused by ice shelf thinning, gradual ice shelf front retreat or ice shelf disintegration, has been linked to 17 instantaneous ice velocity increases and thus dynamic thinning since the early 1990s. This link is clearly 18 evident in the Amundsen and, to a lesser degree, Bellingshausen sectors (Gudmundsson et al., 2019), where 19 passive shelf ice (ice that can be removed without major effects on the ice shelf dynamics) is very limited or 20 absent (Fürst et al., 2016). Surface mass balance (SMB) changes, dominated by snowfall, exhibit strong 21 regional and temporal variability, for example with multidecadal increases in the Antarctic Peninsula 22 inferred since the 1930s (Medley and Thomas, 2019), and dominate the interannual to decadal variability of 23 the AIS mass balance (Rignot et al., 2019). However, no significant continent-wide SMB trend is inferred 24 since 1979 (The IMBIE Team et al., 2018; Medley and Thomas, 2019) (regional changes of Antarctic SMB 25 are assessed further in Atlas Section 11.1). In summary, there is very high confidence that the observed AIS 26 mass loss since the early 1990s is primarily linked to ice shelf changes. 27 28 29 [START FIGURE 9.18 HERE] 30 31 Figure 9.18: (a) A p-box (Section 9.6.3.2) based estimate of the range of values of paleo Antarctic ice sheet mass and 32 sea level equivalents relative to present day and the median over all central estimates (Bamber et al., 33 2009; Argus and Peltier, 2010; Dolan et al., 2011; Mackintosh et al., 2011; Golledge et al., 2012; Miller 34 et al., 2012b; Whitehouse et al., 2012; Golledge et al., 2013; Ivins et al., 2013; Argus et al., 2014; Briggs 35 et al., 2014; Golledge et al., 2014; Maris et al., 2014; De Boer et al., 2015; Dutton et al., 2015; Golledge 36 et al., 2015; Pollard et al., 2015; DeConto and Pollard, 2016; Gasson et al., 2016; Goelzer et al., 2016; 37 Yan et al., 2016; de Boer et al., 2017; Golledge et al., 2017b; Kopp et al., 2017; Simms et al., 2019) ; and 38 cumulative mass loss (and sea level equivalent) since 2015, with satellite observations shown from 1993 39 (Bamber et al., 2018a; The IMBIE Team et al., 2018; WCRP Global Sea Level Budget Group, 2018a) 40 and observations from 1979 (Rignot et al., 2019), ISMIP6 projected changes by 2100 under 41 RCP8.5/SSP5-8.5 and RCP2.6/SSP1-2.6 scenarios (thin lines from (Seroussi et al., 2020; Edwards et al., 42 2021; Payne et al., 2021) and 17th to 83rd, 5th to 95th percentile ranges of the ISMIP6 emulation (shaded 43 line, (Edwards et al., 2021)). Right, 17th to 83rd, 5th to 95th percentile ranges for ISMIP6, emulator, and 44 LARMIP-2 including SMB at 2100. Schematic interpretations of individual reconstructions (Anderson et 45 al., 2002; Bentley et al., 2014; De Boer et al., 2015; Goelzer et al., 2016) of the spatial extent of the 46 Antarctic ice sheet are shown for the (b) mid-Pliocene Warm Period, (c) the Last Interglacial and (d) the 47 Last Glacial Maximum (Fretwell et al., 2013): grey shading shows extent of grounded ice. Maps of mean 48 elevation changes (e) 1978-2017 derived from multi-mission satellite altimetry (Schröder et al., 2019) and 49 (f) ISMIP6 (2061-2100) projected changes for an ensemble using the NorESM1-M climate model under 50 the RCP8.5 scenario (Seroussi et al., 2020). Further details on data sources and processing are available 51 in the chapter data table (Table 9.SM.9). 52 53 [END FIGURE 9.18 HERE] 54 55 56 The SROCC stated with high confidence that melting of ice shelves by warm ocean waters, leading to 57 reduction of ice shelf buttressing, has driven the observed ongoing thinning of major WAIS outlet glaciers. Do Not Cite, Quote or Distribute 9-65 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Since the SROCC, digitized radar measurements have shown that the eastern ice shelf of Thwaites Glacier in 2 the Amundsen Sea Embayment thinned between 10 and 33% during the three decades after 1978 (Schroeder 3 et al., 2019), and the role of basal ice shelf melting has been emphasized (Smith et al., 2020). Strong surface 4 meltwater production has been noted as a precursor of ice shelf disintegration in and since the SROCC (Bell 5 et al., 2018), and recent work placed strong meltwater production events (Lenaerts et al., 2017; Nicolas et al., 6 2017; Wille et al., 2019) and seasons (Robel and Banwell, 2019) in this context. Antarctic ice shelf basal 7 meltwater flux varied between about 1100±150 Gt yr-1 in the mid-1990s and about 1570±140 Gt yr-1 in the 8 late 2000s before decreasing to 1160 ± 150 Gt yr–1 in 2018, and basal melt rates strongly vary with 9 geographical position and depth, as a function of the surrounding water temperature (Adusumilli et al., 10 2020). Section 9.2.2.3 assesses that the intrusion of warm Circumpolar Deep Water, which has warmed and 11 shoaled since the 1980s, has been at least partially controlled by forcing with significant decadal variability. 12 Limited evidence suggests that beyond strong internal decadal wind variability, increased greenhouse gas 13 forcing has slightly modified the mean local winds between 1920 and 2018, facilitating intrusion of 14 Circumpolar Deep Water heat on the Amundsen-Bellingshausen continental shelf and increased ice-shelf 15 melt (Section 9.2.2.3). However, theoretical understanding is still incomplete and in situ measurements 16 within the ice-ocean boundary layer are sparse (Wåhlin et al., 2020). Moreover, modelling and therefore 17 attribution of ice-shelf basal melt remains challenging because of insufficient process understanding, 18 required spatial resolution, the paucity of in-situ observations (Dinniman et al., 2016; Asay-Davis et al., 19 2017; Turner et al., 2017), and uncertainties of bathymetric datasets under ice shelf cavities (Goldberg et al., 20 2019, 2020; Morlighem et al., 2020). In summary, ice shelf thinning, mainly driven by basal melt, is 21 widespread around the Antarctic coast and particularly strong around the WAIS (high confidence), although 22 basal melt rates show substantial spatio-temporal variability. 23 24 Satellite observations suggest that changes in sea ice coverage and thickness can modulate iceberg calving, 25 ice-shelf flow and glacier terminus position around Antarctica (Miles et al., 2013, 2016, 2017; Massom et al., 26 2015; Greene et al., 2018; Bevan et al., 2019), either through mechanical coupling or via changes to ocean 27 stratification, influencing basal melting. A combined observational and modelling study (Massom et al., 28 2018) showed that regional loss of a protective sea-ice buffer played a role in the rapid disintegration events 29 of the Larsen A and B and Wilkins ice shelves in the Antarctic Peninsula between 1995 and 2009, by 30 exposing damaged (rifted) outer ice-shelf margins to enhanced flexure by storm-generated ocean swells. One 31 observational study (Sun et al., 2019) suggests that the absence of sea ice in front of ice shelves, which leads 32 to strengthened topographic waves, favours ice shelf basal melt rates by increasing the baroclinic (depth 33 varying) ocean heat flux which can enter the cavity (Wåhlin et al., 2020). Paleo evidence for sea ice control 34 on ice sheets is lacking, but geologic evidence shows a concordance between periods of ice sheet growth and 35 the expansion of sea ice (Patterson et al., 2014; Levy et al., 2019), both being favoured by reduced sea 36 surface temperatures. Modelling confirms that sea ice controls the strength of ice mélange (Robel, 2017; 37 Schlemm and Levermann, 2021) and thus influences ice shelf flexure and calving rates and stability of 38 floating ice margins, but one model shows this had negligible effect on AIS retreat rates during past warm 39 periods (Pollard et al., 2018). Loss of ice shelf-proximal sea ice is also associated with increased solar 40 heating of surface waters and increased sub-shelf melting (Bendtsen et al., 2017; Stewart et al., 2019). In 41 summary, although in some cases sea-ice decrease and glacier and ice shelf flow and terminus position 42 changes can have the same common cause, there is medium confidence that sea ice decrease ultimately 43 favours the mass loss of nearby ice shelves through a variety of processes. 44 45 The SROCC stated with high confidence that ice shelf disintegration has driven dynamic thinning in the 46 northern Antarctic Peninsula over recent decades and expressed high confidence in currently ongoing mass 47 loss from glaciers that fed now disintegrated ice shelves, although the mass loss rate has decreased in the 20 48 years since the immediate speed-up following ice shelf disintegration in 1995 and 2002. Observed flow 49 speed of these tributary glaciers is still 26% higher than before the ice-shelf disintegration (Seehaus et al., 50 2018). Conversely, flow speed increase of the tributary glaciers of the Scar Inlet Ice Shelf has been 51 interpreted as a sign of evolving instability of the currently intact ice shelf in one study (Qiao et al., 2020). 52 53 Ongoing grounding line retreat, indicating dynamic thinning, is observed with high confidence in many areas 54 of Antarctica and particularly on the WAIS, with the highest rates being in the Amundsen and 55 Bellingshausen Sea areas, and around Totten Glacier in East Antarctica, as stated in the SROCC. Research Do Not Cite, Quote or Distribute 9-66 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 published since the SROCC has evidenced grounding line retreat of the West Antarctic Berry Glacier on 2 Getz Coast (Millan et al., 2020) and on the East Antarctic Denman Glacier (Brancato et al., 2020), both since 3 1996. Furthermore observed grounding line retreat in excess of 1.5 km between 2003 and 2015 has been 4 reported for parts of Marie Byrd Land (Christie et al., 2018). In summary, there is high confidence that 5 grounding lines of marine-terminating glaciers are currently retreating in many areas around Antarctica, and 6 particularly around the WAIS, and additional areas of grounding line retreat have been evidenced since the 7 SROCC. 8 9 The SROCC stated with medium confidence that sustained mass losses of several major glaciers in the 10 Amundsen Sea Embayment (ASE) are compatible with the onset of Marine Ice Sheet Instability (MISI), but 11 that whether unstable WAIS retreat had begun or was imminent remained a critical uncertainty. New 12 publications since the SROCC have not substantially clarified this question. On the one hand, a study 13 combining satellite measurements with a numerical model with prescribed ice shelf thinning (Gudmundsson 14 et al., 2019) suggests that MISI is not required to explain the observed current mass loss rates of the WAIS, 15 because they are consistent with external climate drivers. Furthermore, the fast grounding-line retreat of the 16 Pine Island Glacier in the ASE, which was triggered in the 1940s (Smith et al., 2017), observed after 1992 17 (Rignot et al., 2014) and previously interpreted as a sign of MISI (Favier et al., 2014), seems to have 18 stabilized recently (Milillo et al., 2017; Konrad et al., 2018), and its current flow patterns do not suggest 19 ongoing or imminent MISI (Bamber and Dawson, 2020). On the other hand, sustained fast grounding line 20 retreat has been observed for Smith Glacier in the ASE (Scheuchl et al., 2016), and an analysis of flow 21 patterns and grounding line retreat of the ASE Thwaites Glacier between 1992 and 2017 (Milillo et al., 2019) 22 showed sustained, albeit spatially heterogeneous, grounding line retreat, highlighting ice-ocean interactions 23 that lead to increased basal melt. In addition, Denman Glacier in East Antarctica was shown to hold potential 24 for unstable retreat (Brancato et al., 2020). In summary, the observed evolution of the ASE glaciers is 25 compatible with, but not unequivocally indicating an ongoing MISI (medium confidence). 26 27 The SROCC reported limited evidence and medium agreement for anthropogenic forcing of the observed 28 AIS mass balance changes. As stated in Section 3.4.3.2, there remains low confidence in attributing the 29 causes of the observed mass of loss from the Antarctic ice sheet since 1993 in spite of some additional 30 process-based evidence to support attribution to anthropogenic forcing. 31 32 33 9.4.2.2 Model evaluation 34 35 The AR5 (Church et al., 2013a; Flato et al., 2013) stated that regional climate models and global models with 36 bias-corrected SST and sea ice concentration tended to produce more accurate simulations of Antarctic 37 surface mass balance (SMB) than coupled climate models, and also noted strong climate model temperature 38 biases over the Antarctic, though the latter may reflect known biases in the reanalysis used (Fréville et al., 39 2014). Section Atlas.11.1 assesses that there is medium confidence in the capacity of climate models to 40 simulate Antarctic climatology and surface mass balance changes. 41 42 Section 9.2.3.2 assesses that there is low confidence in simulations of Southern Ocean temperature. Few 43 ocean models resolve ice shelf cavities, and biases in present-day melt rates can be substantial in some 44 sectors, including the key region of the Amundsen Sea (FESOM: Figure 9.19) (Naughten et al., 2018). An 45 increasing number of observational studies from which basal melt rates are calculated (Huhn et al., 2018; 46 Adusumilli et al., 2020; Das et al., 2020; Hirano et al., 2020; Stevens et al., 2020), combined with improved 47 understanding of water-mass-specific influences and modes of melting or dissolving (Silvano et al., 2018; 48 Adusumilli et al., 2020; Malyarenko et al., 2020; Wåhlin et al., 2020), may help to refine these models in the 49 future. However, given the limited number of available models and their biases, there is currently low 50 confidence in the sub-shelf melt rates simulated by ocean models. 51 52 Improvements in the representation of grounding line evolution in ice sheet models since the AR5 (such as 53 sub-grid schemes for basal friction and ice shelf melt, and local grid refinement) means that most of the 54 model simulations presented in the SROCC were dominated by physical processes. Since then, these 55 advances have been applied in several model intercomparison projects (MISMIP+: Cornford et al., (2020); Do Not Cite, Quote or Distribute 9-67 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 ABUMIP: Sun et al., (2020); ISMIP6 and LARMIP-2: Box 9.3). All models participating in ISMIP6 and 2 LARMIP-2, for example, simulate ice shelf and grounding line evolution, and include subshelf melt 3 parameterisation, which was not the case in the SeaRISE intercomparison (Bindschadler et al., 2013; 4 Nowicki et al., 2013). Simulations of grounding line evolution (Seroussi et al., 2017, 2020) have benefited 5 from improved bedrock topography (Morlighem et al., 2020). Treatment of subshelf melting, however, 6 remains one of the causes of large differences in Antarctic ice sheet models, particularly for partially floating 7 grid cells in models with coarse resolution (Levermann et al., 2020a; Edwards et al., 2021). Due to the 8 limitations in resolving cavities in ocean models, as described above, basal melt rates are generally 9 parameterized at the ice-shelf base, based on ocean model simulations of temperatures and salinity instead 10 (Nowicki et al., 2020b; Seroussi et al., 2020). While this has the advantage of connecting melt rates to 11 emission scenarios, a large variety of melt parameterisations exist (DeConto and Pollard, 2016; Lazeroms et 12 al., 2018; Reese et al., 2018; Hoffman et al., 2019; Pelle et al., 2019; Jourdain et al., 2020), and there is low 13 agreement due to limited observational constraints (ocean temperature, salinity, velocity, and ice shelf 14 draft)(Jourdain et al., 2020), uncertainty in the physics of parameterized processes, missing processes (e.g., 15 tides), and uncertainty in the treatment of ice-sheet–climate feedbacks (Donat-Magnin et al., 2017; 16 Bronselaer et al., 2018; Golledge et al., 2019a). Parameterisations are usually calibrated to present-day melt 17 rates, but can respond differently to projected ocean warming (Favier et al., 2019; Jourdain et al., 2020). Two 18 different calibrations were used in ISMIP6 (Jourdain et al., 2020; Nowicki et al., 2020b;Box 9.3): one 19 reproducing melt rates averaged around the whole continent (MeanAnt: Figure 9.19), and the other 20 reproducing melt rates near the grounding line of Pine Island Glacier (PIGL: Figure 9.19), leading to large 21 differences in melt rates. Evaluation with observations and two cavity-resolving models suggests that the 22 MeanAnt parameterisation better reproduces observed melt rates and projected increases in both the warm 23 Amundsen Sea Embayment and cold Ronne-Filchner shelf cavity, as well as total Antarctic melting 24 (Jourdain et al., 2020). The PIGL calibration represents the upper end for increased basal melt sensitivity that 25 would be caused by continent-wide changes to ocean water properties and circulation under strong future 26 forcing (Jourdain et al., 2020). The basal sliding law also has a strong influence on grounding line retreat and 27 glacier acceleration in response to perturbations, and varies spatially (Sun et al., 2020). Sliding laws (Joughin 28 et al., 2019) can only be constrained with observations in regions experiencing significant change and with 29 sufficiently long observational records. 30 31 32 [START FIGURE 9.19 HERE] 33 34 Figure 9.19: Ice shelf basal melt rates for present-day (upper panels) and changes from present-day to 35 the end of the 21st century under the RCP8.5 scenario (lower panels). Present-day melt 36 rates were estimated through: the input-output method constrained by satellite observations and 37 atmosphere/snow simulations (Rignot et al., 2013) and representative of 2003-2008 (upper left); 38 the non-local-PIGL parameterization constrained by observation-based ocean properties 39 (Jourdain et al., 2020) and representative of 1995-2014 (upper centre); the Finite Element Sea- 40 ice/ice-shelf Ocean Model (FESOM) simulation over 2006-2015, forced by atmospheric 41 conditions from a CMIP5 multi-model mean (MMM) under the RCP8.5 scenario ((Naughten et 42 al., 2018) upper right). Future anomalies are calculated as 2081-2100 minus present-day using 43 the ISMIP6 non-local-MeanAnt and non-local-PIGL parameterizations (Jourdain et al., 2020) 44 lower left and centre respectively) based on projections from the NorESM1-M CMIP5 model, 45 and the FESOM-MMM projection (lower right). Note the symmetric-log colour bar (linear 46 around zero, logarithmic for stronger negative and positive values). Inset highlights the 47 Amundsen Sea Region. Further details on data sources and processing are available in the chapter data 48 table (Table 9.SM.9). 49 50 [END FIGURE 9.19 HERE] 51 52 53 The SROCC noted that Antarctic ice sheet simulations are increasingly evaluated or formally calibrated with 54 modern observations and/or paleodata – to obtain more realistic initial conditions (ice sheet geometry, 55 velocity and forcing) and to constrain uncertainty in probabilistic projections. This trend continues (Nias et Do Not Cite, Quote or Distribute 9-68 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 al., 2019; Gilford et al., 2020; Hamlington et al., 2020b; Wernecke et al., 2020). However, while the large- 2 scale characteristics of the initial ice sheet state have improved significantly (Box 9.3), capturing the smaller- 3 scale rates of change, including mass trends, remains challenging for many models (Goldberg et al., 2015; 4 Reese et al., 2020; Seroussi et al., 2020; Siegert et al., 2020). This increases uncertainty in projections, 5 especially for the 21st century (Section 9.4.2.5). Uncertainties in ice sheet model simulations have been, 6 however, much better quantified since the AR5, through model intercomparison projects (in particular, 7 ISMIP6 and LARMIP-2: Box 9.3), perturbed parameter ensembles, and increasing use of statistical 8 emulation (Gilford et al., 2020; Levermann et al., 2020a; Wernecke et al., 2020; DeConto et al., 2021; 9 Edwards et al., 2021) to better sample the parameter space. By exploring uncertainties more fully, these 10 methods have the potential to identify better simulations of the historical period. 11 12 An important difficulty is how to evaluate simulations of processes that are not currently observed, or rare, 13 or indirectly deduced: in particular, the ice shelf disintegrations and cliff failures that would drive the 14 proposed Marine Ice Cliff Instability (MICI: Section 9.4.2.4 and Box 9.4)(DeConto and Pollard, 2016; 15 DeConto et al., 2021). Models of ice cliff failure can only be indirectly and partially evaluated, using 16 existing (i.e., static) cliffs and laboratory experiments (Clerc et al., 2019). The SROCC stated that there was 17 low agreement on the exact MICI mechanism and limited evidence of its occurrence in the present or the 18 past, and that the validity of MICI remains unproven. Only one ice sheet model represents MICI (Pollard et 19 al., 2015; DeConto and Pollard, 2016; DeConto et al., 2021). The mechanism has not been found to be 20 essential for reproducing Mid Pliocene Warm Period and Last Interglacial reconstructions or satellite 21 observations, though Last Interglacial data slightly favours it in this model (Edwards et al., 2019a; Gilford et 22 al., 2020; DeConto et al., 2021). 23 24 In summary, there is now medium confidence in many ice sheet processes in ice sheet models, including 25 grounding line evolution. However, there remains low confidence in the ocean forcing affecting the basal 26 melt rates and low confidence in simulating mechanisms that have the potential to cause widespread, 27 sustained and very rapid ice loss from Antarctica through MICI. 28 29 30 9.4.2.3 Drivers of future Antarctic ice sheet change 31 32 Surface mass balance 33 The AR5 projected a negative contribution from Antarctic surface mass balance (SMB) changes to sea level 34 over the 21st century (i.e., mitigating sea-level rise), due to increased snowfall associated with warmer air 35 temperatures. Sensitivity of SMB to Antarctic surface air temperature change varied from 3.7 to 7 % ºC-1, 36 and the sea level projections assumed a sensitivity of 5.1 ± 1.5 % ºC-1 from CMIP3 era models (Gregory and 37 Huybrechts, 2006) to estimate SMB changes from Antarctic temperatures in the CMIP5 ensemble. Since the 38 AR5, analyses of CMIP5 and CMIP6 models have found Antarctic temperature sensitivity for accumulation 39 (precipitation minus sublimation) of 3.5 to 8.7 % ºC-1 (Frieler et al., 2015), for SMB of 6.0 to 9.9 % ºC-1 40 (Previdi and Polvani, 2016) and for precipitation of around 4 to 9% ºC-1 (Bracegirdle et al., 2020) (± 1 s.d. 41 ranges). An accumulation sensitivity estimate derived from ice core data lies in the middle of the range ~6% 42 ºC-1 (Frieler et al., 2015). These are consistent, within uncertainties, with each other and the AR5, under the 43 approximation that SMB is dominated by snowfall. 44 45 The AR5 found that the median and likely sea level contributions due to SMB from 1986-2005 to 2100 were 46 -0.05 (-0.09 to -0.02) under RCP8.5 and -0.02 (-0.05 to 0.00) m under RCP2.6. The SROCC did not present 47 a separate SMB contribution, instead showing total Antarctic projections derived from ice sheet models 48 (Section 9.4.2.5). Projections of the SMB contribution to sea level tend to be slightly more negative since the 49 AR5, due at least in part to the higher range in equilibrium climate sensitivity values in CMIP6 (Payne et al., 50 2021). Mean and ± 1 s.d. ranges for grounded Antarctic ice sheet SMB changes from 2000 to 2100 computed 51 from CMIP5 models are -0.08 (-0.13 to -0.04) m SLE for RCP8.5 and similarly for CMIP6 models are -0.07 52 (-0.11 to -0.03) m for SSP5-8.5 (Gorte et al., 2020). The GCMs used to drive ice sheet models in ISMIP6 53 (Box 9.3) project mean grounded AIS SMB changes from 2005 to 2100 of -0.06 (range -0.08 to -0.03) m 54 SLE under RCP8.5 for the six CMIP5 models (Seroussi et al., 2020) and -0.09 (range -0.10 to -0.07) m SLE 55 under SSP5-8.5 for the four CMIP6 models, which have climate sensitivity values of 4.8-5.3 ºC (Payne et al., Do Not Cite, Quote or Distribute 9-69 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 2021). We apply the AR5 parametric AIS SMB model (Section 9.6.3.2) to updated projections of global 2 mean temperature from a two-layer energy budget emulator (Supplementary Material 7.SM.2.), which gives 3 a median -0.05 (5-95% range -0.07 to -0.02) m SLE for SSP5-8.5 (Section 9.4.2.5, Table 9.3), i.e., similar to 4 the AR5 assessment and slightly smaller than the CMIP6 estimate. This estimate is used to augment the 5 LARMIP-2 dynamic projections (Box 9.3) in Sections 9.4.2.5 and 9.4.2.6. Overall, then, CMIP5 and CMIP6 6 GCM simulations of sea level fall by 2100 due to Antarctic SMB increases are around 2-4 cm greater than 7 estimates derived with the statistical method used in the AR5. Further details about projections of Antarctic 8 temperature, precipitation and SMB are provided in Section Atlas.11.1.4, which assesses that, due to the 9 challenges of model evaluation (Section 9.4.2.2) and the possibility of increased meltwater runoff (Kittel et 10 al., 2021), there is only medium confidence that the future contribution of Antarctic SMB to sea level this 11 century will be negative under all greenhouse gas emissions scenarios. Longer timescales are discussed in 12 9.4.2.6. 13 14 Sub-shelf melting 15 The SROCC highlighted that an important ongoing deficiency in projections of Antarctic sub-shelf melting 16 is the lack of ice-ocean coupling in most continental-scale studies. Increased basal melting is mainly caused 17 by warmer Circumpolar Deep Water (CDW; Section 9.2.2.3) on the continental shelves and warming surface 18 waters intruding under ice shelves (Naughten et al., 2018). Predicting whether or not open ocean water 19 masses will freely penetrate ice-shelf cavities, or will be partially blocked by ocean density gradients, is 20 complex (Wåhlin et al., 2020), and whilst melting related to CDW inflow is currently dominant in the 21 Amundsen Sea Embayment, melt in other embayments is limited by deep inflows of high salinity shelf water 22 or seasonally-warmed shallow incursions of Antarctic Surface Water (Stewart et al., 2019; Adusumilli et al., 23 2020). There is little consensus regarding future change in CDW (Section 9.2.2.3), and more generally low 24 confidence in future change in the temperature of Antarctic ice shelf cavities (Section 9.2.3.2). 25 26 The response of sub-shelf melting to ocean warming is also poorly constrained. A key unknown is whether, 27 and when, cold ice shelf cavities might become more similar to the Amundsen Sea Embayment, not only in 28 ocean temperature but also ice-ocean heat exchange, which depends on the cavity geometry and ocean 29 circulation (Little et al., 2009). Only two ocean models with ice shelf cavities have been used to make 30 subshelf basal melting projections for SRES and RCP scenarios (Hellmer et al., 2012; Timmermann and 31 Hellmer, 2013; Timmermann and Goeller, 2017; Naughten et al., 2018). FESOM forced by a CMIP5 multi- 32 model mean under RCP8.5 projects a 90% increase in melting (Figure 9.19), although this could be 33 overestimated due to an underestimation of present day melt rates (Naughten et al., 2018)(Section 9.4.2.2). 34 The temperature-melt relationship was parameterised by ISMIP6 in terms of heat exchange velocity in m a-1, 35 and by LARMIP-2 as basal melt sensitivity in m a-1 ºC-1 (Box 9.3; (Jourdain et al., 2020; Levermann et al., 36 2020a; Reese et al., 2020), and both vary widely around the continent depending on cavity type. Median 37 values of ISMIP6 heat exchange velocity vary by a factor of 5-10 when calibrating to either mean Antarctic 38 or high Pine Island Glacier observed melt rates (Section 9.4.2.2; Box 9.3; (Jourdain et al., 2020). Basal melt 39 sensitivities near the grounding line estimated by Reese et al., (2020) with a box model of ocean overturning 40 range from 3.9 m a-1 ºC-1 for the Weddell Sea to 10.5 m a-1 ºC-1 for the Amundsen Sea region, with a 41 continental mean of 5.3 m a-1 ºC-1. Similarly high Amundsen Sea sensitivities are estimated in coupled ice- 42 ocean simulations of Thwaites Glacier (mean 9.4 m a-1 ºC-1; range 6 to 16 m a-1 ºC-1)(Seroussi et al., 2017). 43 These large variations lead to large differences in basal melt rates and projected sea level contributions when 44 applied to the whole ice sheet in ISMIP6 and LARMIP-2 (Box 9.3). Projections of melt rates from the two 45 ISMIP6 calibrations are higher than those from FESOM driven by a CMIP5 multi-model mean (Figure 9.19; 46 Jourdain et al., 2020). The ISMIP6 ensemble mostly uses the mean Antarctic calibration, but includes some 47 simulations with the Pine Island Glacier calibration and the ISMIP6 emulator samples more of these higher 48 values; LARMIP-2 use basal melt sensitivities (7 to 16 m a-1 ºC-1) consistent with estimates for the 49 Amundsen Sea Embayment. Due to the limited availability of cavity-resolving ocean models and the wide 50 regional variation in estimates of basal melt sensitivity to ocean temperature, there is therefore only low 51 confidence in projected future sub-ice shelf melt rates. The impact of this uncertainty on Antarctic ice sheet 52 model projections to 2100 is discussed in Section 9.4.2.5. 53 54 Ice shelf disintegration 55 Antarctic ice shelves modulate grounded ice flow through buttressing, so their weakening or disintegration is Do Not Cite, Quote or Distribute 9-70 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 crucial for the timing and magnitude of ice loss and onset of instabilities (Section 9.4.2.4; Box 9.4). 2 Projections of ice shelf disintegration are uncertain in terms of both atmospheric warming and the response 3 of the shelf surface (surface melting, and whether shelves then disintegrate due to hydrofracturing and 4 flexing, or are resilient through refreezing or drainage; (Bell et al., 2018). The SROCC stated it is not 5 expected that widespread ice shelf loss will occur before the end of the 21st century, but this was based on 6 only one study, using a regional climate model forced by five GCMs (Trusel et al., 2015), so there was low 7 confidence in this assessment. The study of DeConto and Pollard (2016) projected the appearance of 8 extensive surface meltwater several decades earlier than (Trusel et al., 2015) and was therefore assessed to 9 be too uncertain to include in the SROCC projections of the Antarctic ice sheet. 10 11 Since the SROCC, further studies have highlighted the modelling uncertainties in this area. Coastal surface 12 air temperature projections in CMIP6 models show large inter-model differences driven by sea ice retreat, 13 and exhibit more warming relative to global mean temperature under low emissions than high, due to 14 delayed response of the Southern Ocean to stabilised emissions and stratospheric ozone recovery 15 (Bracegirdle et al., 2020). The updated study of DeConto et al., (9998) includes improvements to the climate 16 simulations relative to those in DeConto and Pollard, (2016) and the resulting surface meltwater projections 17 are now consistent with Trusel et al., (2015). However, the net effect of meltwater feedbacks on ice shelves 18 is uncertain. Ice discharge is expected to lead to surface ocean and atmosphere cooling: this increases ocean 19 stratification and sub-shelf melting, but also reduces ice shelf surface melting and delays hydrofracturing 20 (Golledge et al., 2019a; Sadai et al., 2020; DeConto et al., 2021). The new studies are insufficient to change 21 the SROCC low confidence assessment on ice shelf loss. The consequence of this uncertainty on projections 22 is discussed in Section 9.4.2.5 and Box 9.4. 23 24 25 9.4.2.4 Ice sheet instabilities 26 27 A major uncertainty in future Antarctic mass losses is the possibility of rapid and/or irreversible ice losses 28 through instability of marine parts of the ice sheet, proposed via the mechanisms of Marine Ice Sheet 29 Instability (MISI) and Marine Ice Cliff Instability (MICI), and whether these processes will lead to a collapse 30 of the West Antarctic ice sheet (WAIS). 31 32 MISI is a proposed self-reinforcing mechanism within marine ice sheets that lie on a bed that slopes down 33 towards the interior of the ice sheet, whereby, in the absence of ice shelf buttressing, the position of the 34 grounding line is inherently unstable until reaching an upward sloping bed. The SROCC (Meredith et al., 35 2019) noted advances in modelling MISI since the AR5, but that 'significant discrepancies' remained in 36 projections due to poor understanding of mechanisms and lack of observational data to constrain the models. 37 Since the SROCC, modelling uncertainties have been more thoroughly explored, rather than constrained. 38 (compatibility of current observations in the Amundsen Sea Embayment with MISI is assessed in Section 39 9.4.2.1). Internal climate variability might either slow (Hoffman et al., 2019) or amplify (Robel et al., 2019) 40 MISI, and stable grounding line positions can be reached on downward sloping beds if ice shelves provide 41 buttressing (Sergienko and Wingham, 2019; Cornford et al., 2020). Ice sheet model simulations that remove 42 all Antarctic ice shelves (and prevent them from reforming) show 2-10 m SLE Antarctic mass loss after 500 43 years due to MISI, of which WAIS collapse contributes 2–5 m (Sun et al., 2020), with the majority of the 44 mass loss in the first one to two centuries. Much of the multi-model variation is due to the sliding law 45 (Section 9.4.2.2). However, it is not expected that widespread ice shelf loss will occur before the end of the 46 21st century (Section 9.4.2.3; Box 9.4). A recent update of bed topography that has unveiled large and 47 overdeepened subglacial troughs in East Antarctica potentially vulnerable to MISI (Morlighem et al., 2020) 48 has only been used by a few models (Seroussi et al., 2020; Sun et al., 2020), so current projections could 49 underestimate vulnerability in these regions. The sea level rise contribution of the Antarctic ice sheet 50 therefore crucially depends on the behaviour of individual ice shelves and outlet glacier systems and whether 51 they enter MISI for a given level of warming (Pattyn and Morlighem, 2020, Box 9.4). As for Antarctic 52 simulations generally (Sections 9.4.2.2, 9.4.2.3), there is medium confidence in simulating MISI but low 53 confidence in projecting the subshelf melting and ice shelf disintegration that drive it. 54 55 The SROCC noted limited evidence from geological records and ice sheet modelling suggesting that parts of Do Not Cite, Quote or Distribute 9-71 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 the AIS experienced rapid (centennial) retreat likely due to MISI between 20,000 and 9,000 years ago, and 2 also described more uncertain evidence for the LIG and MPWP. Recent support for past MISI is provided by 3 model simulations of the WAIS during the LIG (Clark et al., 2020), the British Ice Sheet during the last 4 termination (Gandy et al., 2018) and the Laurentide Ice Sheet during the Younger Dryas (Pico et al., 2019), 5 which show progressive retreat despite declining temperatures, indicative of a true (ice dynamic) instability. 6 Direct observational evidence of rapid paleo-ice sheet grounding-line retreat is rare, but on the Larsen 7 continental shelf retreat rates of >10 km yr-1 during the deglaciation have been estimated (Dowdeswell et al., 8 2020). MISI has also been inferred from sedimentological evidence of ice loss from Wilkes Subglacial 9 Basin, East Antarctica (Bertram et al., 2018; Wilson et al., 2018; Blackburn et al., 2020) but these 10 reconstructions cannot unambiguously identify unstable from progressive retreat. Therefore there is limited 11 evidence to identify the operation of instability mechanisms such as MISI in paleo-ice sheet retreat. 12 13 The SROCC assessed that ice-sheet interactions with the solid Earth are not expected to substantially slow 14 sea-level rise from marine-based ice in Antarctica over the 21st century (medium confidence), but that these 15 processes could become important on multi-century and longer time scales. More recent modelling of 16 deglaciation of the Ross Embayment by (Lowry et al., 2020) is consistent with this assessment. However, 17 new projections for Pine Island Glacier (Kachuck et al., 2020) support previous work (Barletta et al., 2018) 18 suggesting lower mantle viscosity in this region leads to a negative feedback on decadal time scales. 19 Grounding-line stabilisation by the solid Earth response may therefore occur over the 21st century in the 20 Amundsen Sea Embayment, where most mass loss is occurring (Section 9.4.2.1), but more generally occurs 21 over multi-centennial to millennial timescales (medium confidence). 22 23 The MICI hypothesis describes rapid, unmitigated calving triggered by ice shelf collapse (Pollard et al., 24 2015). The SROCC noted that the MICI mechanism led one model (DeConto and Pollard, 2016) to lose 25 mass far more rapidly, but excluded the mechanism from its projections due to uncertainty in the timing of 26 the ice shelf disintegration (Section 9.4.2.3). They stated that MICI could lead to sea level contributions 27 beyond 2100 considerably higher than the likely range projected by other models, but given the low 28 agreement on the exact MICI mechanism and limited evidence of its occurrence in the present or the past 29 (Section 9.4.2.2), its potential to affect future sea level rise was very uncertain. Since the SROCC, new 30 simulations show later ice shelf disintegration, in agreement with other models (DeConto et al., 9998; 31 Section 9.4.2.3), and therefore lower projections at 2100 (Section 9.4.2.5). New theoretical evidence 32 suggests that ice cliff collapse may only occur after very rapid ice shelf disintegration caused by unusually 33 high meltwater production (Clerc et al., 2019; Robel and Banwell, 2019), and that the subsequent rate of 34 retreat depends on the terminus geometry (Bassis and Ultee, 2019). As SROCC noted, only Crane Glacier on 35 the Peninsula has shown retreat consistent with MICI, after the Larsen B ice shelf collapsed, and MICI-style 36 behaviour at Jakobshavn and Helheim glaciers in Greenland might not be representative of wider Antarctic 37 glaciers. Observations from Greenland show that steep cliffs commonly evolve into short floating 38 extensions, rather than collapsing catastrophically (Joughin et al., 2020). As assessed in Section 9.4.2.2 and 39 9.4.2.3, there is therefore low confidence in simulating mechanisms that have the potential to cause 40 widespread, sustained and very rapid ice loss from Antarctica this century through MICI, and low confidence 41 in projecting the driver of ice shelf disintegration. 42 43 In summary, poorly understood processes of instabilities, characterized by deep uncertainty, have the 44 potential to strongly increase Antarctic mass loss under high greenhouse gas emissions on century to 45 multicentury timescales (Box 9.4). These instabilities are therefore considered separately in assessments of 46 the future contribution to GMSL (Sections 9.4.2.5, 9.4.2.6, 9.6.3.2, 9.6.3.5). 47 48 49 9.4.2.5 Projections to 2100 50 51 The AR5 assessed the median and likely (66-100% probability) sea level contributions of the Antarctic ice 52 sheet (AIS) in 2100 relative to 1986-2005 to be 0.06 (-0.04 to 0.16) m SLE under RCP2.6 and 0.04 (-0.08 to 53 0.14) m SLE under RCP8.5 (Table 9.3; no change when using the AR6 baseline). The AR5 stated that only 54 the collapse of the marine-based sectors of the AIS, if initiated, could cause global mean sea level to rise 55 substantially above the likely range during the 21st century, with medium confidence this would not exceed Do Not Cite, Quote or Distribute 9-72 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 several tenths of a metre during this period. The assessment of the dynamical contribution had no 2 dependence on emissions scenario, due to the lack of literature, so the decrease in sea level contribution in 3 the higher emissions scenario was solely due to increased SMB (Section 9.4.2.3). The SROCC 4 (Oppenheimer et al., 2019) assessed the total contribution based on five new ice sheet modelling studies that 5 incorporated marine ice sheet dynamics, combining their estimates and interpreting the 5-95th percentile 6 range of the resulting distribution as the likely range (17-83% probability interval, i.e., not open-ended as in 7 the AR5). The median and likely range contributions by 2100 were 0.04 (0.01–0.11) m under RCP2.6 and 8 0.12 (0.03-0.28) m under RCP8.5 (Table 9.3). The positive scenario-dependence in the SROCC - where 9 increases in dynamic losses driven by ocean warming and ice shelf disintegration under higher emissions 10 (Section 9.4.2.3) dominate over increases in surface mass balance - arose from a combination of physical 11 processes and model limitations. Modelling improvements in these studies included improved 12 representations of grounding line response to drivers, more extensive exploration of uncertainties, and 13 inclusion of a positive feedback of meltwater on climate (Golledge et al., 2019b). However, two of the 14 projections did not include surface mass balance changes that would offset dynamic losses (Levermann et 15 al., 2014; Ritz et al., 2015), and the scenario dependence may have been further amplified by highly sensitive 16 subshelf melt parameterisations and use of simplified surface mass balance schemes (Golledge et al., 2015, 17 2019b; Bulthuis et al., 2019; Oppenheimer et al., 2019). 18 19 Since the SROCC, new projections have arisen from multi-model intercomparison projects ISMIP6 and 20 LARMIP-2 (Box 9.3), with one model including MICI (Section 9.4.2.4) (DeConto et al., 2021)(Table 9.3). 21 Corrections are added to allow comparison: all ISMIP6-derived projections have an estimate of the historical 22 dynamical response to pre-2015 climate forcing added, which increases contributions (Box 9.3; Figure 9.18); 23 the LARMIP-2 dynamic projections are combined with an estimate of surface mass balance, which decreases 24 contributions (Sections 9.4.2.3, 9.6.3.2); and the ISMIP6 emulated and LARMIP-2 projections were re- 25 estimated using the global surface air temperature distributions from the two-layer energy budget emulator 26 described in Supplementary Material 7.SM.2. The majority of the new projections indicate that the AIS will 27 overall lose mass and contribute to sea level rise, under all emissions scenarios. Most thinning occurs in the 28 Amundsen Sea sector in WAIS and Totten Glacier in EAIS (Figure 9.18). The most negative contribution is 29 -0.02 m (5th percentile of ISMIP6 combined RCP8.5 and SSP5-8.5 projections after correction) and the 30 largest contribution is 0.57 m SLE (95th percentile; (Levermann et al., 2020b)), or 0.63 m SLE with MICI 31 (95th percentile; (DeConto et al., 2021)). ISMIP6 ensemble ranges are wider for the high scenarios 32 (RCP8.5/SSP5-8.5) than the low (RCP2.6/SSP1-2.6), in part because more simulations were available. The 33 ISMIP6 simulations that apply an ice shelf collapse scenario based on exceedance of a surface meltwater 34 threshold (Trusel et al., 2015) driven by CMIP5 models show only a small increase in mass loss (~0-0.04 m), 35 mostly from the Peninsula, due in part to the small number of ice shelves predicted to collapse this century 36 (Seroussi et al., 2020). Simulations driven by the CMIP5 model HadGEM2-ES, which has unusually extreme 37 warming in the Ross Sea (Barthel et al., 2020b), show a larger mass loss (up to ~0.05 m) in East Antarctica 38 under ice shelf collapse (Edwards et al., 2021). The ISMIP6 projections do not include the efficient 39 meltwater drainage or atmospheric feedbacks that could reduce mass loss further (Seroussi et al., 2020). 40 41 The relationship between emissions scenario and AIS response varies across the studies, with emulated 42 ISMIP6 projections showing a slight negative scenario-dependence in the median (-0.01 m) from SSP1-2.6 43 to SSP5-8.5, and LARMIP-2-based projections showing a slight positive scenario-dependence in the median 44 (0.02 m) (Table 9.3). A lack of clear scenario-dependence in the median masks large individual variations 45 across climate and ice sheet models, whereby the net AIS contribution response to emissions scenario 46 depends on the relative magnitudes of the atmosphere, ocean and ice sheet responses (Barthel et al., 2020b; 47 Seroussi et al., 2020; Edwards et al., 2021). Climate and ice sheet models do not project that the AIS 48 response will be the same under high or low greenhouse gas emissions in 2100, but rather there is no 49 consensus on the sign of the change. In contrast, strong scenario dependence is seen from RCP4.5 to RCP8.5 50 in projections that allow MICI (Section 9.4.2.4;(DeConto et al., 2021), though less so than earlier projections 51 (DeConto and Pollard, 2016) due to later ice shelf disintegrations. A negative or positive scenario- 52 dependence of the AIS response this century cannot be deduced from recent observations, because there is 53 still low confidence in attributing the causes of observed mass loss (Section 9.4.2.1), and neither regional 54 mass increases by surface mass balance nor regional mass losses by ice flow have a linear relationship with 55 global mean temperature (Sections 9.4.2.1, 9.4.2.2, 9.4.2.3). There is therefore low agreement on the Do Not Cite, Quote or Distribute 9-73 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 relationship between emissions scenario and AIS response. However, in the longer-term, mass loss is 2 expected to dominate (Section 9.4.2.6). 3 4 The LARMIP-2 median projections are higher than those of the ISMIP6 emulator (by 0.04-0.07 m), and the 5 95th percentiles are two to three times higher. Two possible reasons for the differences between the emulated 6 ISMIP6 and LARMIP-2 projections are assessed: the set of ice sheet models (Annex II) and the parameter 7 values determining subshelf melt sensitivity to ocean temperature (Section 9.4.2.3; Box 9.3). Using only the 8 thirteen ice sheet models common to ISMIP6 and LARMIP-2 reduces the LARMIP-2 median projections by 9 0.02-0.03 m SLE and the 95th percentiles by 0.04-0.08m SLE (Table 9.3), approximately halving the 10 difference in medians but having relatively small effect on the upper end. Subshelf melt sensitivity has a 11 larger effect, due to the wide variation of estimates from different regions and methods. Using only the Pine 12 Island Glacier subshelf melt distribution (Sections 9.4.2.2, 9.4.2.3) in the ISMIP6 emulator gives a median 13 Antarctic projection of ~0.08 m in 2100 in all scenarios before historical correction, compared with ~0 m 14 using only the mean Antarctic distribution; the published projections use a joint distribution (Edwards et al., 15 2021). Reese et al., (2020) find that using the basal melt sensitivities of LARMIP-2 yield an order of 16 magnitude greater mass loss under RCP8.5 than with the ISMIP6 mean Antarctic values. Halving the basal 17 melt sensitivity parameter range (i.e., in line with a continental mean estimate: Section 9.4.2.3) would lead to 18 a halving of the LARMIP-2 dynamic contribution. This would reconcile the LARMIP-2 and ISMIP6 19 emulator median and 95th percentile projections using the common subset of models within ~0.02-0.05 m. 20 There is therefore limited evidence that the ISMIP6 and LARMIP-2 projections could be reconciled by using 21 common ice sheet models and basal melt sensitivity values. 22 23 It is not possible to distinguish which of ISMIP6 and LARMIP-2 is more realistic due to limitations in 24 historical simulations (Box 9.3) and understanding of basal melting (Section 9.4.3), so the projections are 25 combined using a 'p-box' approach (Section 9.6.3.2). The mean of the ISMIP6 emulated and LARMIP-2 26 medians gives the assessed median projections, and the outer edges of the 17-83% ranges give the outer 27 edges of the assessed likely (17-83%) ranges, i.e., encompassing the structural and parametric uncertainties 28 of both methods, giving medium confidence in their combined projections. The main difference between this 29 assessment and the SROCC is to increase the medians of the lower scenarios by 0.05-0.07 m, so that all SSPs 30 are similar to the SROCC assessment of RCP8.5, and to substantially increase the upper ends of the likely 31 ranges: by 0.14-0.16 m for RCP2.6/SSP1-2.6 and RCP4.5/SSP2-4.5, and 0.06 m for RCP8.5/SSP5-8.5. The 32 increase relative to the SROCC is partly due to the increase in LARMIP-2 projections relative to the original 33 LARMIP study (Levermann et al., 2014), arising from the larger number of participating ice sheet models 34 (Levermann et al., 2020b). The historical dynamic response to pre-2015 climate forcing applied to the 35 ISMIP6 emulator could be overestimated, due to the assumption of a constant future rate (Box 9.3). This 36 assessment encompasses the SROCC and all projections since, except the 83rd percentiles of projections that 37 allow MICI under RCP8.5 (DeConto et al., 2021) and the structured expert judgement under 5C shown in 38 the SROCC (Bamber et al., 2019). Both are used in further p-box estimates to give the outer limits of low 39 confidence assessments (Section 9.6.3.2). 40 41 In summary, it is likely that the Antarctic ice sheet will continue to lose mass throughout this century under 42 all emissions scenarios, i.e., that dynamic losses driven by ocean warming and ice shelf disintegration will 43 likely continue to outpace increasing snowfall (medium confidence). The upper end of projections is not well 44 constrained, due to different assumptions about the future sensitivity of subshelf basal melting to ocean 45 warming and the proposed Marine Ice Cliff Instability triggered by ice shelf disintegration (Sections 9.4.2.3 46 and 9.4.2.4; Box 9.4). 47 48 49 [START TABLE 9.3 HERE] 50 51 Table 9.3: Projected sea level contributions in meters from the Antarctic ice sheet in 2100 relative to 1995-2014, 52 unless otherwise stated, for selected RCP and SSP scenarios. Italics denote partial contributions. The 53 historical dynamic response omitted from ISMIP6 simulations is estimated to be 0.33 ± 0.16 mm yr-1 54 (0.03 m ± 0.01 m in 2100 relative to 2015; Box 9.3). The climate forcing is described in Supplementary 55 Material 7.SM.2. Do Not Cite, Quote or Distribute 9-74 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Representative Concentration Pathways (RCPs) Study RCP2.6 RCP4.5 RCP8.5 Notes IPCC AR5 (Church 0.06 (-0.04 to 0.05 (-0.05 to 0.04 (-0.08 to Median and likely et al., 2013a) 0.16) 0.15) 0.14) (>= 66% range) contribution. IPCC SROCC 0.04 (0.01 to 0.06 (0.01 to 0.12 (0.03 to 0.28) Median and likely (Oppenheimer et al., 0.11) 0.15) (66% range) 2019) contribution. Combination of five studies. ISMIP6 CMIP5- -0.01 to 0.16 --- -0.08 to 0.30 Range of ISMIP6 forced (Seroussi et multi-model al., 2020); excludes contributions in historical dynamic 2100 relative to response 2015 from 2 ESMs for RCP2.6 and 6 ESMs for RCP8.5. LARMIP-2; excludes 0.13 (0.07 to 0.14 (0.07 to 0.17 (0.09 to 0.36) Median (67% range) surface mass balance 0.24) 0.28) [0.06 to 0.58] [90% range] (Levermann et al., [0.04 to 0.37] [0.05 to 0.44] LARMIP-2 multi- 2020b) model dynamic contribution in 2100 relative to 1900. MICI 0.08 (0.06 to 0.09 (0.07 to 0.34 (0.19 to Median (66% (DeConto et al., 0.12) 0.11) 0.53) range) [90% 2021) [0.06 to 0.15] [0.07 to 0.15] [0.11 to 0.63] range] Shared Socioeconomic Pathways (SSPs) Study SSP1-2.6 SSP2-4.5 SSP5-8.5 Multi-model ensemble projections ISMIP6 CMIP6- -0.05 to 0.01 --- -0.09 to 0.11 Range of ISMIP6 forced (Payne et al., multi-model 2021); excludes contributions in 2100 historical dynamic relative to 2015 from response 1 ESM for SSP1-2.6 and 4 ESMs for SSP5- 8.5. ISMIP6 all (CMIP5 0.05 (0.04 to 0.08) --- 0.04 (0.00 to 0.12) Median (66% range) and CMIP6-forced) [0.03 to 0.11] [-0.02 to 0.23] [90% range] including historical contribution from dynamic response ISMIP6 CMIP5- and CMIP5-forced multi- model ensembles, (see caption). Emulated ISMIP6; 0.04 (-0.01 to 0.04 (-0.02 to 0.04 (-0.01 to Median (66% range) excludes historical 0.10) 0.10) 0.09) [90% range] dynamic response [-0.05 to 0.14] [-0.06 to 0.14] [-0.05 to 0.14] contribution in 2100 (Edwards et al., 2021) relative to 2015 from emulator of ISMIP6 used with Chapter 7 climate forcing. Do Not Cite, Quote or Distribute 9-75 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI Emulated ISMIP6 0.09 (0.03 to 0.09 (0.03 to 0.08 (0.03 to 0.14) Emulated ISMIP6, total 0.14) 0.14) [0.00 to 0.18] but relative to 1995- [-0.01 to 0.19] [-0.01 to 0.18] 2014 and including historical dynamic response (see caption) Surface mass balance -0.02 (-0.03 to - -0.03 (-0.04 to - -0.05 (-0.07 to - Median (66% range) 0.01) 0.02) 0.03) [90% range] surface [-0.04 to -0.01] [-0.06 to -0.01] [-0.09 to -0.02] mass balance estimated for the AR5, used to correct LARMIP-2 below. LARMIP-2; excludes 0.15 (0.08 to 0.29 0.17 (0.09 to 0.20 (0.10 to 0.39) Median (66% range) surface mass balance ) [0.05 to 0.44] 0.33) [0.06 to [0.07 to 0.61] [90% range] dynamic 0.49] contribution from LARMIP-2 multi- model method used with Chapter 7 climate forcing. LARMIP-2 subset of 0.14 (0.08 to 0.26) 0.15 (0.08 to 0.17 (0.10 to 0.35) As above, but using models; excludes [0.05 to 0.39] 0.29) [0.05 to [0.06 to 0.54] only the 13 of 16 ice surface mass balance 0.45] sheet models common to both ISMIP6 and LARMIP-2. LARMIP-2 subset of 0.11 (0.05 to 0.24) 0.12 (0.05 to 0.12 (0.05 to 0.30) As above, but models; includes [0.03 to 0.37] 0.26) [0.02 to [0.01 to 0.49] including the surface surface mass balance 0.42] mass balance estimate. LARMIP-2 total 0.13 (0.06 to 0.14 (0.06 to 0.15 (0.05 to 0.34) Median (66% range) 0.27) 0.29) [0.01 to 0.57] [90% range] dynamic [0.03 to 0.41] [0.02 to 0.46] contribution from LARMIP-2 multi- model method used with Chapter 7 climate forcing, including the surface mass balance estimate. This assessment: 0.11 (0.03 to 0.11 (0.03 to 0.12 (0.03 to 0.34) Median (66% range) combination of 0.27) 0.29) [0.00 to 0.57] [90% range] emulated ISMIP6 [-0.01 to 0.41] [-0.01 to 0.46] assessment and LARMIP-2 combining emulated ISMIP6 and LARMIP-2. 1 2 [END TABLE 9.3 HERE] 3 4 5 9.4.2.6 Projections beyond 2100 6 7 The SROCC assessed the median and likely range of Antarctic sea level equivalent contributions at 2300 as 8 0.16 (0.07 – 0.37) m under RCP2.6 and 1.46 (0.60 – 2.89) m under RCP8.5, based on three studies. It was 9 noted that deep uncertainty remained beyond 2100: whilst solid Earth feedbacks could reduce ice loss over Do Not Cite, Quote or Distribute 9-76 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 multi-century timescales, Marine Ice Cliff Instability (MICI; Section 9.4.2.4) might give contributions higher 2 than the likely ranges. The SROCC also presented structured expert judgement (SEJ) projections for 3 comparison (Bamber et al., 2019), which give higher values. Since the SROCC, three studies have made 4 projections to 2300. Rodehacke et al., (2020) assessed two methods for implementing precipitation changes 5 (based on repeating 2071-2100 forcings beyond 2100), which both gave negative projections at 2300 6 because the dynamic response was very small (-0.11 to -0.01 m SLE for RCP2.6; -0.25 to -0.07 m for 7 RCP8.5 forcing). In contrast, simulations forced by 2081-2100 ocean-only projections under RCP8.5/SSP5- 8 8.5 beyond 2100, using two implementations of the ISMIP6 'non-local' basal melt parameterisations (Box 9 9.3; Section 9.4.2.2) and two sliding laws, are all positive (0.08 m to 0.96 m SLE by 2300), though these do 10 not include the negative contribution from surface mass balance changes (Lipscomb et al., 2021). Finally, 11 DeConto et al., (9998) update projections for the MICI hypothesis (Section 9.4.2.4) using the extensions of 12 the RCPs to 2300, and obtain far higher contributions: median (17-83%) ranges of 1.09 (0.71 to 1.35) m SLE 13 under RCP2.6 and 9.60 (6.87 to 13.54) m SLE under RCP8.5. These are larger than previous estimates 14 (DeConto and Pollard, 2016), particularly at the upper end: 0.68 (0.29 to 1.13) m SLE for RCP2.6 and 8.40 15 (7.47 to 9.76) m for RCP8.5 (Edwards et al., 2019b), which can largely be explained by the higher maximum 16 ice-cliff calving rate. LARMIP-2 dynamic projections (Box 9.3) are also estimated under the extended SSPs 17 and corrected with surface mass balance (as in Section 9.4.2.5), giving median (17-83%) ranges of 0.40 18 (0.18–0.78) m SLE at 2300 under SSP1-2.6 and 1.57 (0.68–3.14) m under SSP5-8.5. The longer timescale 19 may invalidate the linear response assumption of LARMIP-2, which neglects any self-dampening or self- 20 amplifying processes. The ranges of projections for 2300 without MICI (Golledge et al., 2015; Bulthuis et 21 al., 2019; Levermann et al., 2020a; Rodehacke et al., 2020; Lipscomb et al., 2021); 'assessed ice-sheet 22 contributions' in Section 9.6.3.5) are -0.14 to 0.78 m SLE under RCP2.6/SSP1-2.6, and -0.27 to 3.14 m SLE 23 under RCP8.5/SSP5-8.5. The lower bounds are the 5th percentile of (Bulthuis et al., 2019) and the lowest 24 mean/median from (Rodehacke et al., 2020), respectively; the upper bounds are the 83% percentiles of the 25 LARMIP-2 estimates. These ranges are wider than the SROCC likely ranges, and more consistent with the 26 SEJ (Bamber et al., 2019). However, projections in which Antarctica contributes much more than the 27 assessed ranges under sustained very high greenhouse gas emissions, i.e., around 7–14 m to GMSL by 2300 28 (DeConto et al., 2021), cannot be ruled out and are taken as a sensitivity case (Section 9.6.3.5; Table 9.11). 29 In summary, there is high confidence that Antarctic mass loss will be greater beyond 2100 under high 30 greenhouse gas emissions than low, but the large range of projections mean we have only low confidence in 31 the likely AIS contribution to GMSL by 2300 for a given scenario. Deep uncertainty remains in the role of 32 Antarctic ice sheet instabilities under very high emissions. 33 34 The West and East Antarctic ice sheets are considered to be tipping elements, i.e., susceptible to critical 35 thresholds. The SR1.5 (Hoegh-Guldberg et al., 2018) assessed that a threshold for WAIS instability may be 36 close to 1.5–2°C (medium confidence), as only RCP2.6 led to long-term projections of less than 1 m 37 (Golledge et al., 2015; DeConto and Pollard, 2016). Based on the agreement of a further study (Bulthuis et 38 al., 2019), the SROCC confirmed that low emissions would limit Antarctic ice loss over multi-century 39 timescales (high confidence), but it was not possible to determine whether this was sufficient to prevent 40 substantial ice loss (medium confidence). Since the SROCC, new studies have revisited this topic (Garbe et 41 al., 2020; Rodehacke et al., 2020; Van Breedam et al., 2020; DeConto et al., 2021; Lipscomb et al., 2021), 42 allowing a more complete assessment along with other studies (Feldmann and Levermann, 2015; Clark et al., 43 2016; Golledge et al., 2017a; Edwards et al., 2019b) and the extension to LARMIP-2 above. The majority 44 project 0–1.3 m SLE on multi-century timescales under scenarios of 1–2°C warming. Projections can 45 increase up to 2 m SLE under high basal melt sensitivity to ocean warming (Section 9.4.2.3)(Lipscomb et al., 46 2021) or MICI (Section 9.4.2.4). On multi-millennial timescales ( 2,000 years), many projections remain 47 below 1.6 m SLE under 1–2°C warming, i.e. less than about half of the West Antarctic ice sheet in sea level 48 equivalent (see also Section 9.6.3.5 and Figure 9.30). Other studies project majority or total loss of WAIS 49 under 1–2°C warming, exceeding 2 m SLE, under the higher end of the warming range ( 1.5°C), or high 50 ocean warming ( 0.5°C) and/or high basal melting around WAIS, or MICI. All but two of these multi- 51 millennial studies use variants of the same ice sheet model, though different modelling choices mean they 52 can be considered quasi-independent. Simulations of previous interglacial periods often show near or total 53 WAIS disintegration, with mass loss exceeding 3 m SLE (e.g. Figure 9.18), although limitations of these 54 studies or inferences that can be drawn under different forcings limit confidence in the robustness of these as 55 quantitative analogues (Sections 9.4.2.4, 9.6.2). Overall, increased evidence and agreement on the timescales Do Not Cite, Quote or Distribute 9-77 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 and drivers of mass loss confirm the SR1.5 assessment that a threshold for WAIS instability may be close to 2 1.5–2°C (medium confidence), and that the probability of passing a threshold is larger for 2°C warming than 3 for 1.5°C (medium confidence), particularly under strong ocean warming. New projections agree with 4 previous studies that only part of WAIS would be lost on multi-century timescales if warming remains less 5 than 2°C (medium confidence). There is limited agreement about whether complete disintegration would 6 eventually occur at this level of warming, but medium confidence this would take millennia. 7 8 Under ~2–3C peak warming, complete or near-complete loss of the West Antarctic ice sheet is projected in 9 most studies after multiple millennia (low confidence), with continent-wide mass losses of ~2–5 m SLE or 10 more; this could occur on multi-century timescales under very high basal melting (Lipscomb et al., 2021) or 11 widespread ice shelf loss and/or MICI (Sun et al., 2020; DeConto et al., 2021) (low confidence). Mass losses 12 under ~2–3C warming could be less than 2 m SLE, particularly for multi-century timescales, low basal 13 melting, or less responsive sliding laws. If warming exceeds ~3C above pre-industrial, part of the East 14 Antarctic ice sheet (typically the Wilkes Subglacial Basin) is projected to be lost on multi-millennial 15 timescales (low confidence), with total AIS mass loss equivalent to around 6–12 m or more sea level rise; 16 mass loss could be much smaller if the dynamic response is small (Bulthuis et al., 2019; Rodehacke et al., 17 2020), or much faster under widespread ice shelf loss and/or MICI (Sun et al., 2020; DeConto et al., 2021). 18 A new study by Garbe et al., (2020) suggests that 6C sustained warming and associated mass loss of ~12 m 19 SLE may be a critical threshold beyond which the ice sheet re-organises to a new state, leading to large 20 losses from East Antarctica (including the Aurora Subglacial Basin) and leading to a further 10 m sea level 21 contribution per degree of warming; other studies also show much higher mass loss per C at higher levels of 22 warming (Van Breedam et al., 2020; DeConto et al., 2021) (Section 9.6.3.5; Figure 9.30). 23 24 The SROCC (Meredith et al., 2019; Oppenheimer et al., 2019) assessed that Antarctic mass losses could be 25 irreversible over decades to millennia (low confidence). Garbe et al., (2020) show that the Antarctic ice sheet 26 is always volumetrically smaller when regrowing under a given warming level than when it retreats under 27 the same forcing, and that even if retreat followed by regrowth results in a net zero change in volume, the 28 spatial distribution of mass may be altered, especially in parts of West Antarctica vulnerable to MISI. 29 Projections that start reducing CO2 concentrations from 2030 onwards, reaching pre-industrial levels around 30 2300, show sea level contributions exceeding 1 m by 2500 when including MICI (DeConto et al., 2021). 31 New research therefore confirms the SROCC assessment that mass loss from the AIS is irreversible on 32 decadal to millennial timescales (low confidence) (FAQ 9.1), and suggests that reducing atmospheric CO2 33 concentrations or temperatures to pre-industrial levels may not be sufficient to prevent or reverse substantial 34 Antarctic mass losses (low confidence). 35 36 37 9.5 Glaciers, permafrost and seasonal snow cover 38 39 9.5.1 Glaciers 40 41 9.5.1.1 Observed and reconstructed glacier extent and mass changes 42 43 Global glacier contribution 44 The AR5 (Vaughan et al., 2013) assessed glacier changes from studies based on the regions defined in the 45 Randolph Glacier Inventory (RGI; version 2.0): a satellite observation-based, global inventory of glacier 46 outlines for the year 2000. Following the SROCC (Hock et al., 2019b; Meredith et al., 2019), we report 47 studies based on RGI version 6.0 (RGI Consortium, 2017). Increased volume of satellite observations and 48 the inclusion of detailed regional glacier inventories has resulted in an improved inventory (RGI Consortium, 49 2017). A new consensus estimate for the ice thickness distribution of all glaciers in RGI 6.0 was obtained 50 from an ensemble of five numerical models (although only one out of five models covered all regions 51 (Farinotti et al., 2019)) calibrated and validated with the worldwide Glacier Thickness Database (GlaThiDa 52 3.0; (GlaThiDa Consortium, 2019; Welty et al., 2020)) where possible. The updated inventory shows 53 decreases in estimated glacier volume in the Arctic, High Mountain Asia and Southern Andes, partially 54 compensated by increases in Antarctica. 15% of the total glacier volume is estimated to be below sea level 55 and would not contribute to sea-level rise if melted (Farinotti et al., 2019). Supplementary Material Table Do Not Cite, Quote or Distribute 9-78 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 9.SM.2 shows the inventory glacier area and mass for each region in the year 2000. 2 3 The SROCC found a globally coherent trend of glacier decline in the last decades despite large annual 4 variability and regional differences (very high confidence). Section 2.3.2.3 assesses the global glacier mass 5 changes for the whole 20th century (see Table 9.5 for contribution to the sea-level budget, note that the 6 peripheral glaciers in Greenland and Antarctica are added to the ice sheets for the budget). The AR6 7 assessment is based on Marzeion et al., (2015), using glacier-length reconstructions (Leclercq et al., 2011) 8 and a glacier model forced by gridded climate observations (Marzeion et al., 2012), and not considering the 9 estimated mass loss of uncharted glaciers (100± 50 Gt yr-1) (Parkes and Marzeion, 2018). The time series are 10 assumed independent resulting in larger uncertainty than presented in the SROCC (see also Section 9.6.1). 11 The rate of global glacier mass loss (excluding the periphery of ice sheets) for the period 1901-1990 is 12 estimated to be very likely 210 ± 90 Gt yr-1, representing 16 [28 to 7]% of the glacier mass in 1901, in 13 agreement with the SROCC within uncertainty estimates. 14 15 Since the SROCC, new regional estimates for the Andes (Dussaillant et al., 2019), High Mountain Asia 16 (Shean et al., 2020), Iceland (Aðalgeirsdóttir et al., 2020), the European Alps (Davaze et al., 2020; Sommer 17 et al., 2020) and Svalbard (Schuler et al., 2020), two new global (Ciracì et al., 2020; Hugonnet et al., 2021) 18 and an ad-hoc estimate for the latest glaciological observations (Zemp et al., 2020) have extended the glacier 19 mass change time series up to 2018/2019 (Figure 9.21 and Supplementary Material Table 9.SM.3). A 20 reconciled global estimate for the period 1962 to 2019 has been compiled by Slater et al., (2021). However, 21 in contrast to Slater et al., (2021), after 2000 this assessment is based on the first globally complete and 22 consistent estimate of 21st-century glacier mass change from differencing of digital elevation models 23 (Hugonnet et al., 2021) covering 94.7% of glacier area with glacier mass change for each glacier in the 24 inventory produced with unprecedented accuracy. The estimates from (Hugonnet et al., 2021) agree within 25 uncertainties with new and previous estimates at global (Hock et al., 2019b; Wouters et al., 2019; Zemp et 26 al., 2019; Ciracì et al., 2020; Slater et al., 2021) and regional scale (Dussaillant et al., 2019; Aðalgeirsdóttir 27 et al., 2020; Schuler et al., 2020; Shean et al., 2020). Excluding peripheral glaciers of ice sheets (RGI regions 28 5 and 19), glacier mass loss rate was very likely 170 ± 80 Gt yr-1 for the period 1971 to 2019 (8 [14 - 4]% of 29 1971 glacier mass), 210 ± 50 Gt yr-1 over the period 1993 to 2019 (6 [8 - 4]% of 1993 glacier mass) and 240 30 ± 40 Gt yr-1 over the period 2006-2019 (3 [4 - 2]% of 2006 glacier mass) (Sections 2.3.2.3, 9.6.1; Table 9.54; 31 Cross-Chapter Box 9.1). Including the peripheral glaciers of the ice sheets, the global glacier mass loss rate 32 in the period 2000-2019 is very likely 266 ± 16 Gt yr-1 (4 [6 -3]% of glacier mass in 2000) with an increase in 33 the mass loss rate from 240 ± 9 Gt yr-1 in 2000-2009 to 290 ± 10 Gt yr-1 in 2010-2019 (high confidence). 34 These estimates are in agreement with the SROCC estimate and extend the period to 2018/19. In summary, 35 new evidence published since the SROCC shows that during the decade 2010 to 2019 glaciers lost more 36 mass than in any other decade since the beginning of the observational record (very high confidence; Figure 37 9.20; Section 8.3.1.7.1). 38 39 40 [START FIGURE 9.20 HERE] 41 42 Figure 9.20: Global and regional glacier mass change rate between 1960 and 2019. The time series of annual and 43 decadal mean mass change are based on glaciological and geodetic balances (Zemp et al. (2019) and 44 Zemp et al. (2020)). Superimposed are the 2002-2019 average rates by (Ciracì et al., 2020) based on the 45 Gravity Recovery and Climate Experiment (GRACE), 2006-2015 estimated rates as assessed in SROCC 46 and the new decadal averages (2000-2009 and 2010-2019) by Hugonnet et al. (2021). (*) New regional 47 estimates for the Andes (Dussaillant et al., 2019), High Mountain Asia (Shean et al., 2020), Iceland 48 (Aðalgeirsdóttir et al., 2020), Central Europe (Sommer et al., 2020) and Svalbard (Schuler et al., 2020) 49 are also shown. The uncertainty reported in each study is shown. See Figure 9.2 for the location of each 50 region. Further details on data sources and processing are available in the chapter data table (Table 51 9.SM.9). 52 [END FIGURE 9.20 HERE] 4 The periods in Table 9.5 end in 2018 leading to a slight difference in the values Do Not Cite, Quote or Distribute 9-79 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Regional glacier changes 2 A major advance since the SROCC is the availability of high accuracy mass loss estimates for individual 3 glaciers (Hugonnet et al., 2021). These results show that during the last 20 years, the highest regional mass 4 loss rates (>720 kg m-2 yr-1) were observed in the Southern Andes, New Zealand, Alaska, Central Europe, 5 and Iceland. Meanwhile, the lowest regional mass loss rates (<250 kg m-2 yr-1) were observed in High 6 Mountain Asia, the Russian Arctic, and the periphery of Antarctica. Glacier mass loss in Alaska (25% of 7 2000-2019 total mass loss), the periphery of Greenland (13%), Arctic Canada North (11%), Arctic Canada 8 South (10%), the periphery of Antarctica (8%), the Southern Andes (8%) and High Mountain Asia (8%), 9 represent the majority (83%) of the total glacier mass loss during the last 20 years (2000-2019). 10 11 The glacier mass loss rate from geodetic mass balance assessments in the Southern Andes during 2006-2015 12 was smaller (720 ± 70 kg m-2 yr-1) (Braun et al., 2019; Dussaillant et al., 2019; Hugonnet et al., 2021) than 13 previously assessed in the SROCC (860 ± 160 kg m-2 yr-1), though within uncertainties. In the Central and 14 Desert regions of the Southern Andes, an increase in mass loss from 2000-2009 to 2010-2018, and a high 15 loss rate in Patagonia for the whole period, are observed (Dussaillant et al., 2019). Records of glacier mass 16 loss in Peru (Seehaus et al., 2019a) and Bolivia (Seehaus et al., 2019b) in the period 2000-2016 show an 17 increase in mass loss towards the end of the observation period. In Western North America, outside of 18 Alaska and western Yukon, there was a fourfold increase in mass loss for 2009-2018 (860 ± 320 kg m-2 yr-1) 19 compared to 2000-2009 (203 ± 214 kg m-2 yr-1) (Menounos et al., 2019), and in the Canadian Arctic there 20 was a doubling of mass loss in the last two decades compared with pre-1996 (Noël et al., 2018; Cook et al., 21 2019). The peripheral glaciers in NE Greenland experienced a 23% increase in mass loss in 1980-2014 22 compared to the period 1910-1978/87 (Carrivick et al., 2019). In Iceland, 16 ± 4 % of the ~1890 glacier mass 23 has been lost; about half of that loss occurred in the period 1994-2019 (Aðalgeirsdóttir et al., 2020). Glacier 24 records starting in 1960 in Norway show that half of the observed glaciers advanced in the 1990s but all have 25 retreated since 2000 (Andreassen et al., 2020). In Svalbard, glaciers have been losing mass since the 1960s 26 with a tendency towards more negative mass balance since 2000 (Deschamps-Berger et al., 2019; Van Pelt et 27 al., 2019; Morris et al., 2020; Noël et al., 2020; Schuler et al., 2020). A similar increase in mass loss has been 28 observed for Franz Josef Land in the Russian Arctic (Zheng et al., 2018). Rapid retreat and downwasting 29 throughout the European Alps in the early 21st century is reported (Sommer et al., 2020) and long term 30 records, although limited, indicate sustained glacier mass loss in High Mountain Asia since ~1850, with 31 increased mass loss in recent decades (Shean et al., 2020). In summary, although interannual variability is 32 high in many regions, glacier mass records throughout the world show with very high confidence that the 33 loss rate has been increasing in the last two decades (see also Section 8.3.1.7.1 and 12.4 for regional glacier 34 assessment). 35 36 Section 2.3.2.3 assesses that the rate and global character of glacier retreat in the latter part of 20th century 37 and finds the first decades of the 21st century appear to be unusual in the context of the Holocene (medium 38 confidence) and the global glacier recession in the beginning of the 21st century to be unprecedented in the 39 last 2000 years (medium confidence). These assessments are supported by regional evidence. New 40 reconstructions of the Patagonian Ice Sheet suggest that 20th-century glacial recession occurred faster than at 41 any time during the Holocene (Davies et al., 2020). The reconstructions of glacier variations show that the 42 glaciers in some regions are now smaller than previously recorded: since the mid-16th century in the Mont 43 Blanc and Grindelwald regions of the European Alps (Nussbaumer and Zumbühl, 2012), since the 9th 44 century in Norway (Nesje et al., 2012), and for the past 1800 years in NW Iceland (Harning et al., 2016, 45 2018). In Arctic Canada and Svalbard, many glaciers are now smaller than they have been in at least 4000 46 years (Lowell et al., 2013; Miller et al., 2013, 2017, Schweinsberg et al., 2017, 2018) and more than 40,000 47 years in Baffin Island (Pendleton et al., 2019). Although the millennial glacier length variation records are 48 incomplete and discontinuous, and glacier fluctuations depend on multiple factors (e.g. temperature, 49 precipitation, topography, internal glacial dynamics), there is a coherent relationship between rising 50 temperatures, negative mass balance and glacier retreat on centennial timescales across most of the world. 51 Glaciological and geodetic observations show that the rates of early 21st-century mass loss are the highest 52 since 1850 (Zemp et al., 2015). For all regions with long-term observations, glacier mass in the decade 2010 53 to 2019 was the smallest since at least the beginning of the 20th century (medium confidence). 54 55 In contrast to the global glacier mass decline (Figure 9.21, Supplementary Material 9.SM.2, Table 9.5), a few Do Not Cite, Quote or Distribute 9-80 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 glaciers have gained mass or advanced due to internal glacier dynamics or locally restricted climatic causes. 2 The SROCC discusses the “Karakoram anomaly” (centred on the western Kunlun range (at about 3 80°E/35°N), but also covering part of the Pamir and Karakoram ranges), where glaciers have been close to 4 balance since at least the 1970s and had a slightly positive mass balance since the 2000s. Since the SROCC, 5 new evidence suggests that this anomaly is related to a combination of low-temperature sensitivity of debris- 6 covered glaciers, a decrease of summer air temperatures (Cross-Chapter Box 10.3), and an increase in 7 snowfall possibly caused by increases in evapotranspiration from irrigated agriculture (Bonekamp et al., 8 2019; de Kok et al., 2020; Farinotti et al., 2020; Shean et al., 2020). However, a recent geodetic mass 9 balance estimate suggests substantially increased thinning rates of High Mountain Asian glaciers after about 10 2010 (Hugonnet et al.,2021). There is limited evidence to assess whether the “Karakoram Anomaly” will 11 persist in coming decades, but due to the projected increase in air temperature throughout the region its long- 12 term persistence is unlikely (high confidence) (Kraaijenbrink et al., 2017; de Kok et al., 2020; Farinotti et al., 13 2020; Cross-Chapter Box 10.3). 14 15 16 Drivers of glacier change 17 The AR5 (Masson-Delmotte et al., 2013) noted that early-to-mid-Holocene glacier minima could be 18 attributed to high summer insolation (high confidence), unlike the current situation. Since the AR5, new and 19 improved chronologies of glacier size variations from the end of the last glacial period and the Holocene 20 (e.g., Solomina et al., 2015, 2016; Eaves et al., 2019; Hall et al., 2019; Marcott et al., 2019; Bohleber et al., 21 2020; Davies et al., 2020; Palacios et al., 2020) confirm the dominant role of orbital forcing for millennial- 22 scale glacier fluctuations but emphasize the role of other forcings – solar and volcanic activity, ocean 23 circulation, sea ice and internal climate variability – in explaining the regional variability of glacier 24 fluctuations at shorter time scales. Shakun et al., (2015) demonstrated that during the last deglacial transition 25 (18-11 ka), the mid-to-low-latitude glacier retreat was driven by an increase in atmospheric CO2 and global 26 temperature. 27 28 In the Northern Hemisphere, where summer insolation decreased during the Holocene (Section 2.2.1), 29 glaciers generally waxed (Briner et al., 2016; Kaufman et al., 2016; Lecavalier et al., 2017; Zhang et al., 30 2017; Axford et al., 2019; Geirsdóttir et al., 2019; Larsen et al., 2019; Luckman et al., 2020). Conversely, in 31 the Southern Hemisphere, where summer insolation increased during the Holocene, glaciers generally waned 32 (Solomina et al., 2015; Kaplan et al., 2016; Reynhout et al., 2019). However, these general global trends 33 were modulated by regional climate variations in temperature and precipitation (Murari et al., 2014; Kaplan 34 et al., 2016; Batbaatar et al., 2018; Saha et al., 2018) and there are a number of examples of this. A 35 precipitation increase led to a local early Holocene (7-8 ka) glacier maximum in arid Mongolia (Gichginii 36 Range). Glacier advances at ~9 ka in Southwest Greenland have been suggested to be a consequence of the 37 freshwater pulse from the Laurentide Ice Sheet, which led to cooling in the Baffin Bay area (Schweinsberg et 38 al., 2018). Lake sediments indicate that the glaciers in the region were smaller than today or absent between 39 8.6 and 1.4 ka BP (Larocca et al., 2020). Glaciers on the Antarctic Peninsula and in Patagonia during the 40 Holocene were strongly affected by the Southern Westerly Winds, sea ice extent, and ocean circulation 41 (García et al., 2020). Recent studies indicate that explosive volcanism can drive glacier advances (Solomina 42 et al., 2015, 2016; Schweinsberg et al., 2018; Brönnimann et al., 2019). In summary, on millennial time 43 scales over the Holocene, there is high confidence orbital forcing drove hemispheric-scale glacier variations, 44 but new studies provide a nuanced picture of responses to a variety of regional-scale forcings. 45 46 Section 3.4.3.1 assesses new attribution studies for glaciers and finds that human influence is very likely the 47 main driver of the global, near-universal retreat of glaciers since the 1990s. The SROCC assessed that it is 48 very likely that atmospheric warming is the primary driver for the global glacier recession. Since the 49 SROCC, a study of glaciers in New Zealand used event attribution to confirm a connection between extreme 50 glacier mass loss years and anthropogenic warming (Vargo et al., 2020). 51 52 The SROCC stated with high confidence that besides temperature, other factors, such as precipitation 53 changes or internal glacier dynamics, have modified the temperature-induced glacier response in some 54 regions. Deposition of a thin layer (<2 cm) of light-absorbing particles (e.g., black carbon, brown carbon, 55 algae, mineral dust or volcanic ash) can exert an important control on glacier mass balance, by decreasing Do Not Cite, Quote or Distribute 9-81 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 surface albedo and thus increasing absorbed shortwave radiation and melt (see also section 7.3.4.3. The 2 SROCC found limited evidence and low agreement that this process has had a significant effect on observed 3 long-term glacier changes. Several studies have shown melt increases due to the deposition of light- 4 absorbing particles (Schmale et al., 2017; Wittmann et al., 2017; Sigl et al., 2018; Di Mauro et al., 2019, 5 2020; Magalhães et al., 2019; Constantin et al., 2020). Conversely, increasingly thick debris cover (>2-5 cm) 6 on retreating glaciers can slow down glacier melt (Pratap et al., 2015; Brun et al., 2016). Although debris 7 covers only about 4-7% of the total glacier area globally (Scherler et al., 2018; Herreid and Pellicciotti, 8 2020), many glaciers are heavily debris-covered in their lower reaches, especially in High Mountain Asia, 9 the Caucasus, the European Alps, Southern Andes and Alaska, resulting in different responses to warming 10 than similar clean-ice glaciers. A shift in regional meteorological conditions driven by the location and 11 strength of the upper level zonal wind has been found to have forced recent high mass loss rates in Western 12 North America (Menounos et al., 2019). High geothermal heat flux areas underneath glaciers and high 13 energy dissipation in the flow of water and ice causes additional mass loss of the glaciers in Iceland 14 (Jóhannesson et al., 2020), accounting for 20% of the mass loss since 1994 (Aðalgeirsdóttir et al, 2020). 15 Glacier lake volume, in front of retreating glaciers, has increased globally by around 48% between 1990 and 16 2018 (Shugar et al., 2020), which can increase both subaqueous melt and calving. In summary, there is high 17 confidence that non-climatic drivers have and will continue to modulate the first-order temperature response 18 of glaciers in some regions. 19 20 21 [START FIGURE 9.21 HERE] 22 23 Figure 9.21: Global and regional glacier mass evolution between 1901 and 2100 relative to glacier mass in 2015. 24 Reconstructed glacier mass change through the 20th century (Marzeion et al., 2015) and observed during 25 1961-2016 (Zemp et al., 2019). Projected (2015-2100) glacier mass evolution is based on the median of 26 three Representative Concentration Pathways (RCPs) emission scenarios (Marzeion et al., 2020). 27 Uncertainties are in all cases the 90% confidence interval. For a better comparison between regions, the 28 maximum relative mass change was set to 200%, although for three regions, the volume changes between 29 1901 and 2015 exceeded that value. For the Low Latitude, New Zealand, and High Mountain Asia 30 glaciers, the changes were larger than 1000%, 350%, and 250%, respectively. See Figure 9.2 for the 31 location of each region. Further details on data sources and processing are available in the chapter data 32 table (Table 9.SM.9). 33 34 [END FIGURE 9.21 HERE] 35 36 37 9.5.1.2 Model evaluation 38 39 Since the AR5, glacier mass projections have been coordinated by the Glacier Model Intercomparison 40 Project (GlacierMIP) (Hock et al., 2019a; Marzeion et al., 2020). The SROCC (Hock et al., 2019b) relied on 41 six global-scale glacier models based on previously published glacier model projections (Hock et al., 2019a), 42 and found with high confidence that glaciers will lose substantial mass by the end of the century but assigned 43 medium confidence to the magnitude and timing of the projected glacier mass loss, because of the simplicity 44 of the models, the limited observations in some regions to calibrate them and the diverging initial glacier 45 volumes. 46 47 Since the SROCC, (Marzeion et al., 2020) projected 21st century global-scale glacier mass changes based on 48 seven global-scale and four regional-scale glacier models (Annex II). All models used the same initial and 49 boundary conditions, forming a more coherent ensemble of projections compared to the SROCC. 50 Nevertheless, challenges remain because of scarcity of glacier thickness, surface mass balance and frontal 51 ablation data for model calibration, but also due to uncertainties in glacier outlines, surface elevations and ice 52 velocities. The global surface mass balance models are of varying complexity, including mass balance 53 sensitivity approaches (van de Wal and Wild, 2001), temperature-index methods (Anderson and Mackintosh, 54 2012; Marzeion et al., 2012; Radić et al., 2014; Huss and Hock, 2015; Kraaijenbrink et al., 2017; Maussion 55 et al., 2019a; Zekollari et al., 2019; Rounce et al., 2020) and simplified energy balance calculations (Sakai 56 and Fujita, 2017; Shannon et al., 2019). Compared to simpler, empirical parameterizations, full energy- Do Not Cite, Quote or Distribute 9-82 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 balance models are not necessarily the most appropriate choice for simulating future glacier response to 2 climate change, even at the local scale (Réveillet et al., 2017, 2018), because of parameter and forcing 3 uncertainties. All models account for glacier retreat and advance but only two models (Anderson and 4 Mackintosh, 2012; Huss and Hock, 2015) include frontal ablation. 5 6 Secondary processes such as debris-cover thickening (e.g., Herreid and Pellicciotti, 2020), albedo changes 7 due to light-absorbing particles (e.g., Magalhães et al., 2019; Williamson et al., 2019), trends of refreezing 8 and water storage in firn (e.g., Ochwat et al., 2021), dynamic instabilities such as surges (e.g., Thøgersen et 9 al., 2019) or glacier collapse (e.g., Kääb et al., 2018), are not represented in global glacier models, resulting 10 in both underestimated and overestimated sensitivity to warming that is currently not possible to quantify. 11 Furthermore, challenges for future projections are caused by the low resolution and high spatial variability at 12 subgrid scale of the precipitation amount provided by GCMs, which requires downscaling to the spatial scale 13 of a glacier (Maussion et al., 2019a; Zekollari et al., 2019; Marzeion et al., 2020). In summary, in agreement 14 with the SROCC, progress in global scale glacier modelling efforts allows medium confidence in the 15 capabilityof current-generation glacier models to simulate the magnitude and timing of glacier mass changes 16 as a response to the climatic forcing. 17 18 19 9.5.1.3 Projections 20 21 The AR5 (Vaughan et al., 2013) and the SROCC (Hock et al., 2019b) stated with high confidence that the 22 world’s glaciers are presently in imbalance due to the warming of recent decades. The observed retreat of 23 glaciers is only a partial response to the already realized warming (Christian et al., 2018), and they are 24 committed to losing considerable mass in the future, even without further change in air temperature (Mernild 25 et al., 2013; Trüssel et al., 2013; Zekollari and Huybrechts, 2015; Huss and Fischer, 2016; Marzeion et al., 26 2018; Jouvet and Huss, 2019). One model estimates that 36 ± 8 % of global glacier mass is already 27 committed to be lost due to past greenhouse gas emissions (Marzeion et al., 2018). Although accumulation 28 and ablation instantly determine the surface mass balance, the glacier geometries adjust to changed 29 atmospheric conditions over a longer time (Zekollari et al., 2020). The adjustment time, often referred to as 30 the response time, is variable from one glacier to another, depending on the glacier geometry (thickness and 31 steepness), surface mass balance and gradient (e.g., Jóhannesson et al., 1989; Harrison et al., 2001; Lüthi, 32 2009; Zekollari et al., 2020). Response time is variable: years for smaller and steeper glaciers (Beedle et al., 33 2009; Lüthi and Bauder, 2010; Rabatel et al., 2013), up to tens or hundreds of years for larger and gentle- 34 sloped glaciers ( e.g., Burgess and Sharp, 2004; Lüthi et al., 2010; Zekollari et al., 2020). The models 35 indicate that the disequilibrium between the glaciers and present atmospheric conditions (1995 to 2014) 36 reduces and then disappears at around year 2070 (Marzeion et al., 2020). There is therefore very high 37 confidence that the disequilibrium of glaciers will persist as warming continues and that glaciers will 38 continue to lose mass for at least several decades because of their lagged response even if global temperature 39 is stabilized. 40 41 The SROCC assessed that global glacier mass loss by 2100, relative to 2015 will be 18% [likely range 11 to 42 25%] for scenario RCP2.6 and 36% [likely range 26 to 47%] for RCP8.5, and that many glaciers will 43 disappear regardless of the emission scenario (very high confidence). Since the SROCC, new results from 44 (Marzeion et al., 2020) have been published (Box 9.3; Figure 9.21, Table 9.4, including peripheral glaciers in 45 Greenland and Antarctica). Glaciers will lose 29,000 [9,000 to 49,000] Gt and 58,000 [28,000 to 88,000] Gt 46 over the period 2015-2100 for RCP2.6 and RCP8.5, respectively (medium confidence), which represents 18 47 [5 to 31] % and 36 [16 to 56] % of their early 21st century mass, respectively (Table 9.4). Within 48 uncertainties, these agree with the SROCC estimates, although with a slightly smaller mass loss due to the 49 inclusion of models with lower sensitivity to changing climate conditions (Marzeion et al., 2020). The 50 greatest source of uncertainty in glacier mass loss until the middle of the 21st century is the disagreement 51 between glacier models, with emissions scenario becoming the dominant cause of uncertainty by the end of 52 the 21st century (Marzeion et al., 2020). 53 54 Although the GlacierMIP projections (Hock et al., 2019a; Marzeion et al., 2020) were forced by RCP 55 scenarios, two global glacier models (Huss and Hock, 2015; Maussion et al., 2019b) were also run with 13 Do Not Cite, Quote or Distribute 9-83 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 GCMs and SSP scenarios (Table 9.4). These results show increased mass loss compared to the RCP forced 2 simulations, although with fewer global glacier models. To enable the glacier contribution to future sea level 3 rise to be estimated under the full range of SSP scenarios (Section 9.6.3.3), the GlacierMIP results are 4 emulated using a Gaussian Process model (Edwards et al., 2021) (Box 9.3, Table 9.4). The emulated 5 projections show a narrower range than the roughly equivalent RCP projections, which may be explained by 6 not accounting for covariance in the regional uncertainties (Marzeion et al., 2020) and by the fact that the 7 emulator caps sea level contribution for each region at the volume above floatation estimated by (Farinotti et 8 al., 2019) (Table 9.SM.2). Comparison of simulated and emulated regional sea level contributions support 9 this explanation. Rates of change and post-2100 projections for the sea level projections are estimated with 10 the AR5 parametric fit (Supplementary Material 9.SM.4.5;(Church et al., 2013a)) applied to the GlacierMIP 11 results (Marzeion et al., 2020), and these are also shown in Table 9.4 for comparison. 12 13 14 [START TABLE 9.4] 15 16 Table 9.4: Projected sea level contributions in meters from global glaciers (including peripheral glaciers in 17 Greenland and Antarctica) by 2100 relative to 2015, for selected RCP and SSP scenarios. 18 Representative Concentration Pathways (RCPs) Study RCP2.6 RCP4.5 RCP8.5 Notes IPCC AR5 and SROCC 0.10 0.12 0.17 Median and likely (Church et al., 2013a; (0.04 – 0.16) (0.06 – 0.19) (0.09 – 0.25) (66% range) Oppenheimer et al., contributions in 2019) 2100 relative to 1995-2014 GlacierMIP 0.094 0.142 0.200 Mean (± 1 s.d. Hock et al. (2019a) (0.069 – 0.119) (107 – 177) (0.156 – 0.240) range) contributions GlacierMIP 0.079 0.119 0.159 Median Marzeion et al., (2020) [0.023 – 0.135] [0.053 – 0.185] [0.073 – 0.245] [90% range] Shared Socioeconomic Pathways (SSPs) Study SSP1-2.6 SSP2-4.5 SSP5-8.5 Notes GlacierMIP 0.111 0.136 0.190 Mean (66% experimental protocol (0.077 to 0.145) (0.096 to 0.176) (0.133 to 0.247) range) [90% (Marzeion et al., 2020) [0.05 to 0.167] [0.07 to 0.201] [0.09 to 0.283] range] using 13 with CMIP6 forcing GCMs and 2 glacier models1 GlacierMIP (Marzeion et 0.102 0.128 0.171 Median (66% al., 2020) with AR5 (0.076 to 0.134) (0.095 to 0.167) (0.124 to 0.224) range) [90% parametric fit: used for [0.059 to 0.154] [0.076 to 0.192] [0.098 to 0.259] range] rates and post-2100 contribution from projections AR5 parametric (Supplementary Material fit to GlacierMIP 9.SM.4.5) ensemble, relative to 1995- 2014 Emulated (Marzeion et 0.080 0.115 0.170 Median (66% al., 2020); (Edwards et (0.059 to 0.101) (0.093 to 0.137) (0.144 to 0.196) range) [90% al., 2021) [0.046 to 0.116] [0.077 to 0.155] [0.124 to 0.218] range] contribution in 2100 relative to Do Not Cite, Quote or Distribute 9-84 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 2015 from emulator of GlacierMIP6 used with Chapter 7 climate forcing 1 1 OGGM (Maussion et al., 2019a) and GloGEM (Huss and Hock, 2015) 2 3 [END TABLE 9.4] 4 5 6 The mass loss rates vary between regions and there are distinctively different patterns between scenarios 7 (Marzeion et al., 2020). The global models agree that regions characterized by relatively little glacier- 8 covered area (Low Latitude, Central Europe, Caucasus, Western Canada and US, North Asia, Scandinavia 9 and New Zealand) will lose nearly all (>80%) glacier mass by 2100 in the RCP 8.5 scenario, but their 10 corresponding contribution to sea-level rise will be small. A study using detailed ice dynamics for the largest 11 glacier of the European Alps, Great Aletsch Glacier, projects 60% of present ice volume will be lost by 2100 12 in RCP2.6 and an almost complete wastage of the ice in RCP8.5 (Jouvet and Huss, 2019). Due to their larger 13 mass, the largest contribution to sea level rise comes from glaciers in the Arctic and Antarctic regions 14 (Antarctic, Arctic Canada, Alaska, Greenland, Svalbard and Russian Arctic), in spite of having the smallest 15 relative mass loss, and it is expected that they will continue to contribute to sea level rise beyond 2100. The 16 regions with intermediate glacier mass (Southern Andes, High Mountain Asia and Iceland) show decreasing 17 mass loss rates for RCP2.6 throughout the 21st century, and increasing rates for RCP8.5 that peak in the mid 18 to late 21st century (Figure 9.21). The peak in mass loss rate followed by reduction is due to both decreasing 19 glacier volume and stabilizing mass balance (Marzeion et al., 2020). Vatnajökull, the largest glacierin 20 Iceland, is projected to lose about 50% of its mass by 2300 in extended RCP4.5 and 80-100% in extended 21 RCP8.5 scenarios (Schmidt et al., 2019). In summary, both global and regional studies agree that glacier 22 mass loss will continue in all regions, with larger mass loss for high emission scenarios (high confidence) 23 (see also Section 8.4.1.7.1). 24 25 In the AR5 and the SROCC, glacier mass loss beyond 2100 was calculated employing a parametric fit to 26 available model simulations. In section 9.6.3.5, that same parametric fit is applied to (Marzeion et al., 2020) 27 projections resulting in complete glacier mass loss at year 2300 under SSP5-8.5 and 40-100% mass loss 28 under SSP1-2.6. (Clark et al., 2016) simulate glacier mass evolution, not including glaciers peripheral to the 29 Antarctic ice sheet, for different warming levels for the next ten thousand years. There is limited evidence 30 and low confidence that at sustained warming levels between 1.5 and 2°C, about 50-60% of glacier mass will 31 remain, predominantly in the polar regions. At sustained warming levels between 2 and 3°C, about 50-60% 32 of glacier mass outside Antarctica will be lost and at sustained warming levels between 3 and 5 °C, 60-75% 33 of glacier mass outside Antarctica will disappear. Based on (Marzeion et al., 2020), there is medium 34 confidence that nearly all glacier mass in low latitudes, Central Europe, the Caucasus, Western Canada and 35 the US, North Asia, Scandinavia and New Zealand will disappear at this high warming level. 36 37 38 9.5.2 Permafrost 39 40 This section focuses on the physical aspects of permafrost (perennially frozen ground) as an element of the 41 climate system, drawing on the assessment of observed global permafrost changes provided in Section 42 2.3.2.5, and treating more specifically model evaluation and projections. The permafrost carbon feedback is 43 assessed in Box 5.1. Section 12.4 of this report provides permafrost information relevant to impacts and risk 44 on regional scales. 45 46 9.5.2.1 Observed and reconstructed changes 47 48 The current extent of the global permafrost region is about 22 ± 3 × 106 km2 (Gruber, 2012). Permafrost 49 underlies about 15% of Northern Hemisphere land and more than 50% of the unglacierized land north of Do Not Cite, Quote or Distribute 9-85 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 60°N (Zhang et al., 1999; Gruber, 2012; Obu et al., 2019). It is also found in high-altitude areas of mountain 2 ranges in both hemispheres (estimated in the SROCC (Hock et al., 2019b) as representing about 27-29% of 3 the global permafrost area (medium confidence) and most unglacierized areas in Antarctica (Vieira et al., 4 2010; Obu et al., 2020)). Ground ice volume in permafrost is variable, reaching up to 90% in syngenetic 5 permafrost deposits (Kanevskiy et al., 2013; Gilbert et al., 2016). The SROCC (Meredith et al., 2019) 6 reported medium confidence in the estimation that Earth’s total perennial ground ice volume is equivalent to 7 2-10 cm of global sea level (Zhang et al., 2000). There is no evidence suggesting that a large part of this 8 volume, if melted, would run off and contribute to global sea level. Therefore, and because of the modest 9 total volume of mobilizable water, the contribution of permafrost thaw to past and future sea-level budgets is 10 usually neglected (see section 9.6.3.2). 11 12 Permafrost changes mostly refer to changes in extent, temperature and active layer thickness. The SROCC 13 (Hock et al., 2019b; Meredith et al., 2019) reported with very high confidence that record high permafrost 14 temperatures at the depth of the zero annual amplitude (the depth about 10 to 20 m below the surface where 15 the seasonal soil temperature cycle vanishes) were attained in recent decades in the Northern circumpolar 16 permafrost region, high confidence that permafrost has warmed over recent decades in many mountain 17 ranges, and overall very high confidence that global warming over the last decades has led to widespread 18 permafrost warming. As reported in the SROCC, the global (polar and mountain) permafrost temperature has 19 increased at 0.29 ± 0.12°C near the depth of zero annual amplitude between 2007 and 2016 (Biskaborn et al., 20 2019). Stronger warming has been observed in the continuous permafrost zone (0.39 ± 0.15°C) compared to 21 the discontinuous zone (0.20 ± 0.10°C), consistent with the fact that near the melting point, a large amount of 22 energy is required for melting the ice (Figure 9.22), and because of the reduced effect of Arctic amplification 23 in more southerly locations (Romanovsky et al., 2017). This is consistent with longer-term Arctic trends 24 from deep boreholes shown in Figure 2.22. Mountain permafrost temperature trends are heterogeneous, 25 reflecting variations in local conditions such as topography, surface type, soil texture and snow cover, but 26 again, generally weaker warming rates are observed in warmer permafrost at temperatures close to 0°C, 27 particularly when ice content is high (e.g., Mollaret et al., 2019; Noetzli et al., 2019; PERMOS, 2019). In 28 summary, strong variability in recent permafrost temperature trends is linked to local conditions, regionally 29 varying temperature trends and the thermal state of permafrost itself, but as discussed in Section 2.3.2.5, 30 there is overall high confidence in the observed increases in permafrost temperature over the past 3 to 4 31 decades throughout the permafrost regions. 32 33 Closer to the surface, the active layer undergoes annual cycles of freeze and thaw. The SROCC reported 34 medium confidence in active layer thickness (ALT) increase as a pan-Arctic phenomenon. Recent evidence 35 presented in Section 2.3.2.5 shows pervasive ALT increase in the European and Russian Arctic in the 21st 36 century and in high elevation areas in Europe and Asia since the mid-1990s. Emergence of a clearer global 37 picture is hampered by (1) uneven distribution of observing sites, (2) substantial variability among the 38 existing sites, strongly influenced by local conditions (soil constituents and moisture, snow cover, 39 vegetation); (3) interannual variability , and (4) thaw settlement in ice-rich terrain (Streletskiy et al., 2017; 40 O’Neill et al., 2019). In summary, in agreement with the SROCC and recent evidence presented in Section 41 2.3.2.5, there is medium confidence that ALT increase is a pan-Arctic phenomenon. 42 43 There is medium confidence that the observed acceleration and destabilization of rock glaciers is related to 44 warming temperatures and increase in water content at the permafrost table in recent decades (Deline et al., 45 2015; Cicoira et al., 2019; Marcer et al., 2019; PERMOS, 2019; Kenner et al., 2020). There is also medium 46 confidence that observed increases in size and frequency of rock avalanches are linked to permafrost 47 degradation in rock walls (Ravanel et al., 2017; Patton et al., 2019; Tapia Baldis and Trombotto Liaudat, 48 2019). In summary, there is medium confidence that mountain permafrost degradation at high altitude has 49 increased the instability of mountain slopes in the past decade. 50 51 The SROCC assessed with high confidence that the extent of subsea permafrost, formed before submersion 52 on Arctic continental shelves during the last deglaciation, is much reduced compared to older studies that 53 had estimated the entire formerly exposed Arctic shelf area to be underlain by permafrost. This is supported 54 by observations (Shakhova et al., 2017) that show rapid thaw of recently submerged permafrost on the East 55 Siberian Shelf . A modelling study (Overduin et al., 2019) estimates that 97% of permafrost under Arctic Do Not Cite, Quote or Distribute 9-86 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 shelves is currently thinning. 2 3 Based on multiple studies, there is medium confidence that widespread retreat of coastal permafrost is 4 accelerating in the Arctic (Günther et al., 2015; Cunliffe et al., 2019; Isaev et al., 2019). There is also 5 consistent evidence of complete permafrost thaw in areas of discontinuous and sporadic permafrost since 6 about 1980 (Camill, 2005; Kirpotin et al., 2011; James et al., 2013; Jones et al., 2016a; Borge et al., 2017; 7 Chasmer and Hopkinson, 2017; Gibson et al., 2018a), but this evidence is geographically scattered. In spite 8 of increasing evidence of landscape changes from site studies and remote sensing, quantifying permafrost 9 extent change remains challenging because it is a subsurface phenomenon that cannot be observed directly 10 (Jorgenson and Grosse, 2016; Trofaier et al., 2017). A modelling study for the Qinghai-Tibet Plateau 11 between the 1960s and the 2000s (Ran et al., 2018) suggests transition from permafrost to seasonally frozen 12 ground over an area of more than 400,000 km2. In summary, there is medium confidence that complete 13 permafrost thaw in recent decades is a common phenomenon in discontinuous and sporadic permafrost 14 regions. In addition, paleoclimatic evidence presented in Section 2.3.2.5 confirms a long-term sensitivity of 15 permafrost extent to climatic variations, although an analysis of North American speleothem records over the 16 last two glacial cycles indicates that this apparent high sensitivity could be a consequence of regional-scale 17 variability (Batchelor et al., 2019). 18 19 There is a lack of formal studies attributing observed permafrost changes (thaw depth, thermal state) or 20 associated landscape changes to anthropogenic forcing. However, the observed Arctic warming has been 21 attributed to anthropogenic forcing (e.g., Najafi et al., 2015) and an obvious physical link exists between 22 ground temperatures (and thus permafrost) and surface air temperatures. Therefore physically consistent and 23 convergent lines of evidence lead to medium confidence in anthropogenic forcing being the dominant cause 24 of the observed pan-Arctic permafrost changes. Added to this, local permafrost change by soil and 25 ecosystem disturbance is induced by increasing human industrial activities in the Arctic (e.g., Raynolds et 26 al., 2014). 27 28 29 9.5.2.2 Evaluation of permafrost in climate models 30 31 As stated in AR5 (Flato et al., 2013), coupled models contributing to CMIP5 showed large inter-model 32 variability of permafrost extent due to deficiencies in reproducing surface characteristics and processes 33 (Koven et al., 2013), particularly thermal properties of the ground and snow. These deficiencies led the 34 SROCC (Meredith et al., 2019) to express only medium confidence in the models’ capacity to correctly 35 project the magnitude of future permafrost changes, in spite of high confidence in the models’ projection of a 36 general thaw depth increase and a substantial loss of shallow permafrost. The SROCC further noted that 37 several types of physical “pulse” disturbances, in particular fire and thermokarst formation, are usually not 38 represented in coupled climate models. This has been discussed in detail in the SROCC, which assessed that 39 there is high confidence that permafrost degradation through fire (Jones et al., 2015; Gibson et al., 2018b) is 40 currently occurring faster in some well-studied regions than during the first half of the 20th century, and 41 medium confidence that thermokarst formation, to which about 20% of the northern permafrost region is 42 vulnerable (Olefeldt et al., 2016), can lead to faster large-scale permafrost degradation in response to climate 43 change. 44 45 Since the SROCC, dedicated modelling of the evolution of ice- and organic-rich permafrost in the northeast 46 Siberian lowlands (Nitzbon et al., 2020) has shown that not representing thermokarst-inducing processes in 47 ice-rich terrain leads to a systematic underestimation of the rapidity and magnitude of permafrost thaw. 48 Simplified inventory-based modelling (Turetsky et al., 2020) points towards similar conclusions. Although 49 these pulse disturbances still need to be represented in CMIP-type models, there have been many new 50 developments to that type of model since CMIP5 and the AR5. Soil freezing and its thermal and hydrological 51 effects are now included in a large number of land-surface modules that are part of the CMIP6 ensemble 52 (Chadburn et al., 2015a; Hagemann et al., 2016; Cuntz and Haverd, 2018; Guimberteau et al., 2018; 53 Yokohata et al., 2020), sometimes allowing for the effects of excess ice (Lee et al., 2014). Improved 54 representation of snow insulation in models has led to more realistic simulated permafrost extents (e.g., 55 Paquin and Sushama, 2015). In a post-CMIP5 ensemble of land-surface models driven by observed Do Not Cite, Quote or Distribute 9-87 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 meteorological conditions (McGuire et al., 2016), inter-model spread was substantially reduced when the 2 ensemble was restricted to models that appropriately represented the effect of snow insulation on the 3 underlying soil (Wang et al., 2016b). More detailed descriptions of high-latitude vegetation characteristics, 4 vegetation dynamics, and snow-vegetation interactions have been included in several models (Chadburn et 5 al., 2015b; Porada et al., 2016; Druel et al., 2017) since the AR5. 6 7 A total soil column depth of at least about 10 m is required to adequately represent the dampening effect of 8 seasonal-scale heat exchanges between shallow and deeper ground, and thus to correctly simulate active 9 layer thickness (Lawrence et al., 2008; Ekici et al., 2015). However, many CMIP6 models still have 10 shallower total soil columns (Burke et al., 2020) and the proportion of models with deeper total soil columns 11 has not increased since CMIP5 (Koven et al., 2013). Another recently identified process, usually not 12 represented in the current (CMIP6) generation of climate models (Zhu et al., 2019), is warming-driven 13 decomposition and burning of organic material that provides strong thermal insulation of underlying ground. 14 Decay of the insulating organic material can lead to increased permafrost thaw, creating a positive feedback 15 loop. 16 17 In spite of the aforementioned structural improvements to many models, the simulated current permafrost 18 extent from available CMIP6 models shows no substantial improvement with respect to CMIP5 (see Figure 19 9.22a). The extent of the region where permafrost is simulated within the top 15 m in the Northern 20 Hemisphere for the 1979-1998 period is characterized by very large scatter in the coupled CMIP5 and 21 CMIP6 historical simulations compared to estimates of the present permafrost extent based on multiple 22 observational lines of evidence (Zhang et al., 1999) and models based on satellite observations and 23 reanalyses (Gruber, 2012; Obu et al., 2019). Outliers with very low simulated permafrost extent are models 24 that have only a very shallow soil column (leading to an underestimate of thermal inertia at depth) and do not 25 take into account soil water phase changes. These inadequacies lead to an overestimate of seasonal thaw 26 depth, exceeding the total thickness of the models’ soil columns (Burke et al., 2020). Excessive simulated 27 permafrost extent can in several cases be traced to insufficient thermal insulation by the winter snow cover 28 (Burke et al., 2020). 29 30 Figure 9.22a also shows that the corresponding land-atmosphere simulations with prescribed observed sea- 31 surface temperatures and sea-ice concentrations, and the land-only simulations with prescribed reanalysis- 32 based meteorological forcing, do not provide an improved simulation of the current permafrost extent, 33 although, by construction, they can be expected to exhibit lower land surface climate biases. This further 34 points to deficiencies in the land modules as the main reason for biases, consistent with conclusions drawn 35 from the analysis of CMIP5 output (Koven et al., 2013), as reported in the SROCC and the AR5. 36 37 In spite of more realistic description of permafrost-related processes in many coupled climate models, the 38 CMIP6 models thus still produce a very scattered ensemble of estimates of current permafrost extent, and 39 there is high confidence that this is strongly linked to deficiencies of the representation of soil processes. 40 Furthermore, current-generation climate models tend to neglect several physical disturbances that can lead to 41 faster permafrost thaw. Because of large uncertainties in the future evolution of these drivers (see SROCC), 42 there is limited evidence that these shortcomings lead to an underestimate of permafrost degradation rates in 43 response to climate change in the CMIP6 ensemble. In summary, there is high confidence that coupled 44 models correctly simulate the sign of future permafrost changes linked to surface climate changes, but only 45 medium confidence in the amplitude and timing of the transient response. 46 47 48 [START FIGURE 9.22 HERE] 49 50 Figure 9.22: Simulated versus observed permafrost extent and permafrost volume change by warming level. a) 51 Diagnosed Northern Hemisphere permafrost extent (area with perennially frozen ground at 15 m depth, or 52 at the deepest model soil level if this is above 15 m) for 1979-1998, for available CMIP5 and CMIP6 53 models, from the first ensemble member of the historical coupled run, and for CMIP6 AMIP 54 (atmosphere+land surface, prescribed ocean) and land-hist (land only, prescribed atmospheric forcing) 55 runs. Estimates of current permafrost extents based on physical evidence and reanalyses are indicated as 56 black symbols (triangle: Obu et al. (2018); star: Zhang et al. (1999); circle: central value and associated Do Not Cite, Quote or Distribute 9-88 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 range from Gruber (2012)). b) Simulated global permafrost volume change between the surface and 3 m 2 depth as a function of the simulated GSAT change, from the first ensemble members of a selection of 3 scenarios, for available CMIP6 models. Further details on data sources and processing are available in the 4 chapter data table (Table 9.SM.9). 5 6 [END FIGURE 9.22 HERE] 7 8 9 9.5.2.3 Projected permafrost changes 10 11 The AR5 (Collins et al., 2013) and the SROCC (Meredith et al., 2019) (based on available CMIP5 output) 12 both expressed high confidence that future pan-Arctic thaw depth will increase and near-surface permafrost 13 extent will decrease under future global warming, and medium confidence in the magnitude of the simulated 14 changes because of model deficiencies and large spread of the results. 15 16 The equilibrium sensitivity of permafrost extent to stabilized global mean warming has been inferred (by 17 constraining CMIP5 output with diagnosed relationships between the observed present-day spatial 18 distribution of permafrost and air temperature) to be about 4.0×106 km2 °C-1 (Chadburn et al., 2017) for 19 GSAT changes with respect to the present below about +3°C. This equilibrium permafrost sensitivity, 20 relevant for assessing long-term permafrost changes at a stabilized warming level, is about 20% higher than 21 the transient centennial-scale near-surface permafrost extent sensitivity (diagnosed from seasonal thaw down 22 to 3 m depth) suggested by direct analysis of CMIP5 output (Slater and Lawrence, 2013). Compared to these 23 and other studies reported in the AR5 and the SROCC (Koven et al., 2013), the recently suggested 24 equilibrium extent sensitivity to GSAT changes of about 1.5×106 km2 °C-1 based on idealized ground 25 temperature modelling (Liu et al., 2021) appears unrealistically low. 26 27 A strong transient temperature sensitivity of the volume of perennially frozen soil in the top 3 m below the 28 surface is consistently suggested by the available CMIP6 models (Figure 9.22b). Relative to the current 29 volume, the transient sensitivity of the modelled permafrost volume in the top 3 m to GSAT changes (with 30 respect to the 1995-2014 average and up to +3°C change, that is, about up to +4°C with respect to pre- 31 industrial levels) is about 25 ± 5 % °C-1 (Burke et al., 2020), but there is only medium confidence in this 32 value and 1s.d. uncertainty range because of the model deficiencies discussed in 9.5.2.2. It is important to 33 note that permafrost loss will not be limited to the top 3 m, with delayed response of deeper permafrost. The 34 simulated transient temperature sensitivity of permafrost volume is slightly stronger in the SSP1-2.6 scenario 35 than in other SSPs because subsurface temperature lag increases with higher atmospheric warming rates, 36 particularly when ground ice melting induces additional delays. 37 38 Due to the role of air temperature as a major driver of permafrost change, the SROCC (Hock et al., 2019b) 39 expressed very high confidence that permafrost in high-mountain regions is expected to undergo increasing 40 thaw and degradation during the 21st century, with stronger consequences expected for higher greenhouse 41 gas emission scenarios. Recently published studies (e.g., Zhao et al., 2019) support this SROCC assessment. 42 43 In summary, based on high agreement across CMIP6 and older model projections, fundamental process 44 understanding, and paleoclimate evidence, it is virtually certain that permafrost extent and volume will 45 shrink as global climate warms. 46 47 48 9.5.3 Seasonal snow cover 49 50 Mean snow cover extent in January and February, the usual months of maximum extent, covers about 45% 51 of the Northern Hemisphere (NH) land surface (more than 45 million km2 over the 1967-2014 period 52 (Estilow et al., 2015)). In contrast, maximum seasonal snow cover in South America, the dominant ice-free 53 land mass in the Southern Hemisphere in terms of seasonal snow cover extent, remains well below 1 million 54 km2 (Foster et al., 2009) or less than 2% of the Southern Hemisphere land surface. 55 Do Not Cite, Quote or Distribute 9-89 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Terrestrial snow cover is characterized via three variables: (1) areal snow cover extent (SCE), (2) the time 2 period of continuous snow cover (snow cover duration (SCD) which reflects snow-on and snow-off dates) 3 and (3) snow accumulation expressed either as snow depth (SD) or snow water equivalent (SWE; the depth 4 of water stored by the snowpack). 5 6 Observed large-scale snow cover changes, their attribution to human activity and their effects on the 7 hydrological cycle are also discussed in Chapter 2 (Section 2.3.2.2), Chapter 3 (Section 3.4.2) and Chapter 8 8 (Section 8.2.3.1) of this report. The role of snow in the global surface albedo feedback is assessed in Section 9 7.4.2.3. The effect of aerosol deposition on snow albedo and associated climate forcing is assessed in Section 10 7.3.4.3. 11 12 13 9.5.3.1 Observed changes of seasonal snow cover 14 15 The AR5 (Vaughan et al., 2013) reported that NH SCE in June very likely decreased by 11.7% [8.8 to 14.6 16 %] per decade over the 1967-2012 period, exceeding the absolute and relative reductions observed in March 17 and April. The AR5 further reported very high confidence that NH March and April SCE decreased over the 18 90 years after 1922. The SROCC only assessed snow cover changes for the Arctic and mountain areas. For 19 the Arctic (north of 60°N), the SROCC (Meredith et al., 2019) expressed high confidence in SCE decreases 20 of -3.5 ± 1.9% per decade in May and -13.4± 5.4% per decade in June, based on a combination of multiple 21 datasets (Mudryk et al., 2017). Concerning mountain snow cover, the SROCC (Hock et al., 2019b) reported 22 with high confidence that mountain snow cover (both in terms of SCE and maximum SWE) has generally 23 declined since the middle of the 20th century at lower elevations. At higher elevations, the SROCC reported 24 medium confidence in generally insignificant snow cover trends where these were available. The large-scale 25 assessment provided in Section 2.3.2.2 of this report reports very high confidence in substantial reductions of 26 NH snow cover extent (particularly in spring) since 1978, and states that there is limited evidence that this 27 decline extends back to the early 20th century. 28 29 Since the SROCC, progress in characterizing seasonal NH snow cover changes has been made through the 30 combined analysis of datasets from multiple sources (surface observations, remote sensing, land surface 31 models and reanalysis products). A recent combined dataset (Mudryk et al., 2020) identified negative NH 32 SCE trends in all months between 1981 and 2018, exceeding -50 × 103 km2yr-1 in November, December, 33 March and May (Figure 9.23a,b). The loss of spring SCE is also reflected in earlier spring snow melt, 34 derived from surface observations (Bulygina et al., 2011; Brown et al., 2017), satellite observations (Wang et 35 al., 2013; Estilow et al., 2015; Anttila et al., 2018), and model-based analyses (Liston and Hiemstra, 2011). 36 There is considerable inter-dataset and regional variability, but the continental-scale trends of spring snow- 37 off dates from these datasets are consistently negative (Brown et al., 2017; Kouki et al., 2019). 38 39 Satellite-derived estimates of NH SCE compiled within the NOAA snow chart Climate Data Record (NOAA 40 CDR) extend back to 1967, providing one of the longest environmental data records from spaceborne 41 measurements (Estilow et al., 2015). Continental trends from these coarse resolution estimates (~200 km) 42 show declining snow cover during the spring period, consistent with surface warming (Hernández-Henríquez 43 et al., 2015; Mudryk et al., 2017). As assessed in Section 2.3.2.2, there is therefore very high confidence that 44 the NH spring SCE has been decreasing since 1978. 45 46 Hemispheric reconstructions with simple snow models and in-situ observations have extended the satellite 47 record to 1922 (Brown and Robinson, 2011), putting the satellite era in historical context. This study, also 48 assessed in the AR5, suggests an increase in North American spring (March-April) SCE from 1915 to about 49 1950, followed by a decrease of the same total magnitude afterwards. In Eurasia, a negative trend in April is 50 visible over the entire 1922-2010 period of record, while in March, a step decrease at about 1985 separates 51 two periods with insignificant trends. Overall, combining March and April, consistency between the 52 continental trends since 1950 and agreement in sign with the NOAA satellite record since 1967 provides 53 high confidence in Northern Hemisphere spring snow cover decrease since about 1950. Analysis of 54 paleoclimate records (Pederson et al., 2011; Belmecheri et al., 2016) suggests that recent snowpack 55 reductions in western North America are exceptional on a millennial timescale (medium confidence). Do Not Cite, Quote or Distribute 9-90 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 2 Recent remote sensing global-scale studies (Hammond et al., 2018; Notarnicola, 2020) report that since 3 2000, snow cover area and/or duration decreased in 78% of global mountain areas (Notarnicola, 2020). Due 4 to the shortness of these records and high spatial variability, they only provide limited evidence in medium 5 agreement that snow cover area and duration changes over that recent period are more consistently negative 6 at higher (>4000 m) than at lower elevations, and do not alter the high confidence in longer-term mountain 7 snow cover decrease at lower elevations since the middle of the 20th century that was already reported in the 8 SROCC. 9 10 As assessed in detail in Chapter 3 (Section 3.4.2), it is very likely that anthropogenic influence contributed to 11 the observed reductions in Northern Hemisphere springtime snow cover since the mid-20th century. The 12 reasons for this assessment are: (1) physical consistency of the observed spring snowpack and surface 13 temperature changes both in observations and models, (2) the strong observed hemispheric and regional 14 spring SCE and SWE trends, and (3) the general attribution of hemispheric temperature changes to human 15 influence. Consistent between multiple observational products and historical climate model simulations, the 16 observed NH SCE sensitivity to NH land (>30°N) warming (Mudryk et al., 2017) is approximately -1.9 × 17 106 km2°C-1 (95% confidence range of ±0.9×106 km2 °C-1 throughout the snow season. 18 19 Compared to numerous studies on spring SCE changes, less attention has been paid to changes in NH snow 20 cover during the onset period in the autumn, a challenging period to retrieve snow information from optical 21 satellite imagery due to persistent clouds and decreased solar illumination at higher latitudes. Positive trends 22 in October and November SCE in the NOAA-CDR (Hernández-Henríquez et al., 2015) are not replicated in 23 other surface, satellite, and model datasets (Brown and Derksen, 2013; Peng et al., 2013; Hori et al., 2017; 24 Mudryk et al., 2017). The positive trends from the NOAA-CDR are also inconsistent with later autumn 25 snow-on dates since 1980 (-0.6 to -1.4 days per decade), based on historical surface observations, model- 26 derived analyses and independent satellite datasets (updated from Derksen et al., 2017). Furthermore, the 27 SCE trend sensitivity to surface temperature forcing in the NOAA-CDR is anomalous compared to other 28 datasets during October and November (Mudryk et al., 2017). There is therefore medium confidence that the 29 NH SCE trend for the 1981-2016 period was also negative during these two months (Mudryk et al., 2020). 30 31 In the low-to-mid latitude (18°S-40°S) South American Andes, a dry-season snow cover decrease of about 32 12% per decade has been reported for the period 1986-2018 (Cordero et al., 2019), linked to ENSO changes 33 dominant in the northern part and an additional influence of poleward migration of the westerly wind zone in 34 the southern part of the study area. Further South, long-term warming has been identified as the dominant 35 cause of observed winter snow cover reduction over the 1972-2016 period at about 53°S in Brunswick 36 Peninsula (Aguirre et al., 2018). 37 38 The AR5 (Hock et al., 2019b) reported on snow water equivalent (SWE) and snow depth (SD) in situ 39 observations mostly from mountain areas, the majority of which showed negative trends over their respective 40 observational periods. However, the AR5 did not provide an assessment of large-scale snow mass changes 41 across the Northern Hemisphere. The SROCC attributed medium confidence to reports of negative SWE 42 trends in the Russian Arctic between 1966 and 2014, and stated that seasonal maximum snow depth trends in 43 the North American Arctic were mostly insignificant and of inconsistent sign. It further attributed medium 44 confidence to gridded products that suggest negative pan-Arctic SWE trends between 1981 and 2016, and 45 high confidence in a general decline of mountain snow mass at lower elevations, albeit with regional 46 variations. 47 48 Since the AR5, the number of global or hemispheric-scale gridded SWE products has substantially increased. 49 A validation and intercomparison (Mortimer et al., 2020) of datasets derived from (1) reanalysis-based 50 products, (2) a combined surface observation – passive microwave remote sensing product and (3) stand- 51 alone passive microwave products has led to better understanding of the strengths and limitations of each. 52 These gridded products consistently identify negative trends in maximum pre-melt SWE over the 1981 – 53 2016 period over both Eurasia and North America (Mudryk et al., 2020); (Figure 9.23c and d). To further 54 constrain SWE uncertainty, Pulliainen et al. (2020) implemented a bias correction based on snow course 55 observations which yielded a current best estimate for the average 1980-2018 March SWE over NH non- Do Not Cite, Quote or Distribute 9-91 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 alpine land north of 40°N of 2938 Gt [likely range 2846-3062 Gt]. Using this method, the bias-corrected 2 GlobSnow3.0 dataset suggests a 4.6 Gt yr-1 decrease of March SWE over this 39-year period across North 3 America, and a negligible trend across Eurasia. These SWE trends are consistent with the continental SCE 4 trends over this period as assessed above, but strong regional and temporal variability only allows medium 5 confidence in the signs and magnitudes of these trends. However, there is high confidence in a general 6 decline of NH spring SWE since 1981 (Section 2.3.2.2). In the longer term (see also Section 2.3.2.2), annual 7 maximum snow depth trends from site measurements confirm mostly negative trends in North America 8 (Kunkel et al., 2016) between 1960/61 and 2014/15 and strong spatial variability in Eurasia (Zhong et al., 9 2018) between 1966 and 2012, with spatial patterns bearing some resemblance to the shorter satellite-based 10 trends reported by Pulliainen et al. (2020). However, over this longer period, the Eurasian measurements 11 (Zhong et al., 2018) exhibit, on average, a positive trend. On the Qinghai–Tibetan Plateau, site 12 measurements between 1961 and 2010 (Xu et al., 2017) suggest a shift from an initial increase of spring 13 snow depth until about 1980 to a decreasing trend afterwards. 14 15 16 [START FIGURE 9.23 HERE] 17 18 Figure 9.23: Observed monthly Northern Hemisphere (a) snow cover trends and (b) anomalies, and (c) snow 19 mass trends and (d) anomalies. From the observation-based ensemble discussed in the text (Mudryk et 20 al., 2020). Trends and anomalies are calculated over the 1981-2018 period. Further details on data sources 21 and processing are available in the chapter data table (Table 9.SM.9). 22 23 [END FIGURE 9.23 HERE] 24 25 26 Concerning the assessment of SWE trends in mountainous regions, the SROCC noted a need for 27 observations spanning several decades because of very strong temporal variability. Moreover, determining 28 SWE trends in mountain regions is challenging because the coarse resolution (typically 25 to 50 km) of 29 gridded SWE products is inadequate in areas of mountainous terrain (Snauffer et al., 2016). Based on a 30 compilation of a large number of studies of SWE trends in mountain regions, the SROCC noted strong 31 regional variations, but a general consistency in greater reductions in SWE at lower elevations associated 32 with shifts from solid to liquid precipitation. A recent synthesis of snow observations in the European Alps 33 (Matiu et al., 2021) shows a 1971-2019 seasonal (November to May) snow depth trend of -8.4% per decade, 34 along with negative maximum snow depth and seasonal snow cover duration trends. The trends are stronger 35 and more significant during transitional seasons and at transitional (from no snow to snow) altitudes, and 36 exhibit strong regional variations, consistent with earlier reports for the Swiss and Austrian Alps (Schöner et 37 al., 2019) and the Pyrenees (López-Moreno et al., 2020). 38 39 In summary, since the AR5, intercomparison, dataset blending of gridded products, and bias correction using 40 snow course measurements contributed to an improved estimate of the average 1980-2018 March SWE over 41 NH non-alpine land north of 40°N of 2938 Gt [likely range 2846-3062 Gt], with medium confidence in the 42 magnitudes of continental-scale trends over that period. However, there is high confidence in a general 43 decline of NH spring SWE since 1981 (Section 2.3.2.2). In mountain areas, in situ observations tend to 44 suggest that annual maximum SWE reductions are generally stronger at elevation bands where shifts from 45 solid to liquid precipitation affected the snow mass. 46 47 48 9.5.3.2 Evaluation of seasonal snow in climate models 49 50 Building on the AR5 (Flato et al., 2013) and subsequent published work, the SROCC (Meredith et al., 2019) 51 stated that CMIP5 models tended to underestimate the observed decrease of Northern Hemisphere spring 52 snow cover extent due to inappropriate parameterisation of snow processes, misrepresentation of the snow- 53 albedo feedback, underestimated temperature sensitivity, and biased climatological spring snow cover. Since 54 the AR5, progress in the observation, description and understanding of snow microstructure (Kinar and 55 Pomeroy, 2015; Calonne et al., 2017) and its links to physical (thermal and radiative) properties (Löwe et al., Do Not Cite, Quote or Distribute 9-92 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 2013; Calonne et al., 2014) has prompted efforts to represent physical properties as a function of the 2 evolving snow microstructure in models (Carmagnola et al., 2014; Calonne et al., 2015). However, even 3 state-of-the-art snow models intended for meteorological and climate applications still struggle to correctly 4 represent the time evolution of the snow thermal properties, particularly of cold and dry tundra snow 5 (Domine et al., 2016). Moreover, most, if not all, CMIP6 climate models do not explicitly represent the 6 darkening of snow by deposition of BC and other light-absorbing aerosol species that is known to influence 7 snow melt rates (Section 7.3.4.3). Regardless of these shortcomings, snow modules of climate models 8 continue to be improved. Recent progress includes the incorporation of multiple energy balances within the 9 canopy and between subgrid-tiles with different snow heights (Aas et al., 2017; Boone et al., 2017) and 10 inclusion of advanced specific snow models in coupled climate models (Niwano et al., 2018; Voldoire et al., 11 2019), opening the prospect of future progress in quantifying snow-related feedbacks in a changing climate. 12 Recently developed multi-physics snow models (Essery, 2015; Lafaysse et al., 2017), which are able to 13 emulate the behaviour of a large number of models in a broad range of climates, allow model shortcomings 14 and key parameter uncertainties, for example, concerning snow masking by vegetation or snow thermal 15 conductivity, to be identified. Guidance for future model improvement can be provided by improved 16 diagnostics, such as a concise metric of snow insulation (Slater et al., 2017a), which builds on an observed 17 relation between effective seasonal mean snow depth and the dampening of winter season temperature 18 decrease within the soil, and allows an efficient quantification of inaccuracies in the simulated snow 19 insulation effect. 20 21 There is high confidence that large inter-model variations in the snow-cover sensitivity to temperature can 22 largely be explained by inaccuracies in the simulated snow albedo feedback (Qu and Hall, 2014); a multi- 23 model sub-ensemble of CMIP5 models that simulate a correct magnitude of this feedback presents a 40% 24 reduced spread in the projected 21st century Northern Hemisphere land warming trend (Thackeray and 25 Fletcher, 2016). Errors of the simulated feedback strength were linked to 1) systematic positive albedo biases 26 over the boreal forest belt, mostly due to unrealistic treatment of vegetation masking (Thackeray and 27 Fletcher, 2016), 2) inaccurate prescribed tree cover fraction and inappropriate parameterization of leaf area 28 index in some models (Loranty et al., 2014; Wang et al., 2016a) and 3) low spatial resolution leading to 29 inaccuracies in the strength of the simulated SAF in mountainous regions (Letcher and Minder, 2015). 30 Although the representation of snow albedo feedback improved in many CMIP5 models over CMIP3, some 31 models deteriorated (Thackeray et al., 2018). 32 33 Analysis of the available CMIP6 historical simulations for the 1981-2014 shows that on average, CMIP6 34 models simulate well the observed SCE (Mudryk et al., 2020), except for outliers and a median low bias 35 during the winter months (Figure 9.24a). This is an improvement over CMIP5 (Mudryk et al., 2020), in 36 which many snow-related biases were linked to inadequacies of the vegetation masking of snow cover over 37 the boreal forests (Thackeray et al., 2015). A comparison between CMIP5 and CMIP6 results (Mudryk et al., 38 2020) shows that there is no notable progress in the quality of the representation of the observed 1981-2014 39 monthly snow cover trends. 40 41 42 [START FIGURE 9.24 HERE] 43 44 Figure 9.24: Simulated CMIP6 and observed snow cover extent (SCE). a) Simulated CMIP6 and observed 45 (Mudryk et al., 2020) SCE (in millions of km2) for 1981-2014. Boxes and whiskers with outliers 46 represent monthly mean values for the individual CMIP6 models averaged over 1981-2014, with the red 47 bar indicating the median of the CMIP6 multi-model ensemble for that period. The observed interannual 48 distribution over the period is represented in green, with the yellow bar indicating the median. b) Spring 49 (March to May) Northern Hemisphere snow cover extent against GSAT (relative to the 1995-2014 50 average) for the CMIP6 Tier 1 scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5), with linear 51 regressions. Each data point is the mean for one CMIP6 simulation (first ensemble member for each 52 available model) in the corresponding temperature bin. Further details on data sources and processing are 53 available in the chapter data table (Table 9.SM.9). 54 55 [END FIGURE 9.24 HERE] 56 Do Not Cite, Quote or Distribute 9-93 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 9.5.3.3 Projected snow cover changes 2 3 The AR5 (Collins et al., 2013) stated that substantial NH spring snow cover reductions at the end of the 21st 4 century were very likely under strong emission scenarios, and expressed medium confidence in the projected 5 geographic patterns of annual maximum SWE changes. Based on studies using downscaled CMIP5 or RCM 6 output, either directly or via snowpack models driven by such output, the SROCC (Hock et al., 2019b) 7 reported likely snow depth or mass decreases at lower elevations in many mountain ranges over the 21st 8 century and high confidence in smaller future changes at higher elevations. 9 10 Since the AR5, one study (Brown et al., 2017), applying a method developed by de Elía et al. (2013) to a 11 CMIP5 sub-ensemble, suggested that over most of the Northern Hemisphere, the projected decrease of SCD 12 will exceed natural variability before this will be the case for annual maximum SWE. The same study reports 13 that over large parts of Eastern and Western North America and Europe, forced SCD changes are projected 14 to exceed natural variability in the 2020s both in spring and autumn, while the signals tend to emerge later in 15 the Arctic regions and particularly late, after 2060, in Eastern Siberia under the RCP8.5 scenario. Thackeray 16 and Fletcher (2016) have shown that inter-model spread in projected spring SCE trends could be reduced 17 through improved simulation of spring season warming because of the tight coupling between temperature 18 and SCE linked to the snow albedo feedback (Qu and Hall, 2014; Thackeray and Fletcher, 2015). 19 20 Across all emission scenarios and with negligible scenario dependence (Figure 9.24b), CMIP6 models 21 consistently (all models and all months) simulate Northern Hemisphere snow cover decrease in response to 22 future GSAT change over the 21st century (Mudryk et al., 2020). The simulated SCE decrease is close to a 23 linear function of global temperature change for all months (shown in Figure 9.24b for spring, with medium 24 confidence in an average sensitivity of about -8% per °C of GSAT increase), except when snow cover 25 vanishes. This occurs at about +2°C of GSAT change above the 1995-2014 level (that is, about +3°C above 26 the pre-industrial level) for the months of July and August, and at about +3°C above the 1995-2014 level for 27 June and September. Possible effects of such changes on the hydrological cycle are assessed in Chapter 8 28 (Section 8.2.3.1). 29 30 In summary, consistent projections from all generations of global climate models, elementary process 31 understanding and strong covariance between snow cover and temperature on several time scales make it 32 virtually certain that future Northern Hemisphere snow cover extent and duration will continue to decrease 33 as global climate continues to warm, and process understanding strongly suggests that this also applies to 34 Southern Hemisphere seasonal snow cover (high confidence). 35 36 Seasonal snow cover, by definition, has a clear annual cycle with usually complete disappearance in spring 37 and summer and re-formation in autumn or winter. Therefore, there is very high confidence that the current 38 and projected changes to seasonal snow cover are reversible (Verfaillie et al., 2018). In the case of global or 39 regional cooling, abrupt large-scale snow-cover changes with a transition from seasonal to persistent snow 40 cover due to a strong snow albedo feedback are a typical feature of glacial inceptions (e.g., Baum and 41 Crowley, 2003; Calov et al., 2005), and these can be irreversible on centennial or longer timescales because 42 of this feedback. In summary, based on physical understanding and the absence of occurrence of such events 43 in climate model projections, abrupt future changes of seasonal snow cover on large scales in the absence of 44 concomitant abrupt atmospheric change as a driver appear very unlikely in the context of current and 45 projected warming. 46 47 48 9.6 Sea Level Change 49 50 9.6.1 Global and regional sea-level change in the instrumental era 51 52 9.6.1.1 Global mean sea-level change budget in the pre-satellite era 53 54 The SROCC (Oppenheimer et al., 2019) discussed the development and application of new statistical 55 methodologies for reconstructing global mean sea level (GMSL; Box 9.1) from tide-gauge data over the 20th Do Not Cite, Quote or Distribute 9-94 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 century. Based on an ensemble of tide gauge reconstructions, the SROCC assessed an average rate of GMSL 2 rise of 1.38 mm yr-1 [0.81 to 1.95 very likely range] for the period 1901 to 1990. Since the SROCC, two new 3 GMSL reconstructions have been published (Dangendorf et al., 2019; Frederikse et al., 2020b) and are 4 included in an updated ensemble estimate of GMSL change (Palmer et al., 2021, Section 2.3.3.3). Based on 5 these updated data and methods, the GMSL change over the (pre-satellite) period 1901 to 1990 is assessed to 6 be 0.12 m [0.07 to 0.17 very likely range] with an average rate of 1.35 mm yr-1 [0.78 to 1.92 very likely 7 range] (high confidence) (Table 9.5; section 2.3.3.3) in agreement with the SROCC assessment. Both this 8 assessment and SROCC have substantially larger uncertainties than the AR5 assessment, which was based 9 upon a single tide gauge reconstruction and did not account for structural uncertainty (see Palmer et al., 10 (2021) for a discussion). 11 12 The SROCC found that four of the five available tide gauge reconstructions that extend back to at least 1902 13 showed a robust acceleration (high confidence) of GMSL rise over the 20th century, with estimates for the 14 period 1902-2010 (-0.002 to 0.019 mm yr-2) that were consistent with the AR5. New tide gauge 15 reconstructions published since the SROCC (Dangendorf et al., 2019; Frederikse et al., 2020b) support this 16 assessment and suggest that increased ocean heat uptake related to changes in Southern Hemisphere winds 17 and increased mass loss from Greenland are the primary physical mechanisms for the acceleration (section 18 2.3.3.3). Therefore, the SROCC assessment on the acceleration of GMSL rise over the 20th century is 19 maintained. 20 21 The evaluation of the sea-level budget presented here, and in section 9.6.1.2, draws on assessments of the 22 individual components (Section 2.3.3.1 and 9.2.4.1 for global-mean thermosteric and sections 9.5.1.1, 9.4.1.1 23 and 9.4.2.1 for ice mass loss contributions to GMSL change from glaciers and ice sheets). Following the 24 approach of the SROCC, the mass loss from ice sheet peripheral glaciers is included in the ice sheet 25 contributions to GMSL change (glacier mass loss from regions 5 and 19 of the Randolph Glacier Inventory 26 6.0 (RGI Consortium, 2017) are added to ice sheet mass loss where applicable, with uncertainties added in 27 quadrature). The total change in GMSL for each component, and their sum, is summarised in Table 9.5 28 (uncertainties added in quadrature). For consistency across the report and to simplify the treatment of 29 uncertainties, all budget calculations are based on the difference between the first and last year in each period 30 (Palmer et al., 2021), rather than a linear fit to the underlying time series as used in SROCC and AR5. 31 32 The sea-level budget in the SROCC included the anthropogenic contribution of land water storage (LWS; 33 Box 9.1) change from a single estimate (Wada, 2016). Since the SROCC, two studies have combined 34 estimates of natural LWS change with anthropogenic LWS changes from reservoir impoundment and 35 groundwater depletion (Cáceres et al., 2020; Frederikse et al., 2020b). For (Cáceres et al., 2020), zero change 36 is assumed for the period 1901-1948, since their LWS change estimates are not available before 1948. Given 37 the large year-to-year changes associated with hydrological variability, the assessed changes in LWS (Table 38 9.5) are based on linear trends for each period, following (Palmer et al., 2021). Structural uncertainty is 39 estimated from the standard deviation of the trends across the two studies and parametric uncertainty is 40 estimated based on the Monte Carlo simulations of (Frederikse et al., 2020b). These two sources of 41 uncertainty are combined in quadrature and the assessed central estimate is taken as the average of the 42 ensemble mean trends. Compared to the SROCC assessed LWS trend of -0.12 mm yr-1 for the period 1901- 43 1990, the updated assessment leads to a more negative trend of -0.16 [-0.35 to 0.04] mm yr-1, although the 44 two are consistent within the estimated uncertainties. Previous studies and the SROCC have highlighted the 45 large uncertainty in estimates of LWS change over the 20th century (Gregory et al., 2013), and therefore the 46 SROCC assessment of low confidence in the estimated LWS contribution to GMSL change is maintained. 47 48 Since the SROCC, a new ocean heat content reconstruction (Zanna et al., 2019b) (Section 2.3.3.1) has 49 allowed global thermosteric sea-level change to be estimated over the 20th century. As a result, the sea-level 50 budget for the 20th century can now be assessed for the first time. For the periods 1901 to 1990 and 1901 to 51 2018 the assessed very likely range for the sum of components is found to be consistent with the assessed 52 very likely range of observed GMSL change (medium confidence), in agreement with (Frederikse et al., 53 2020b) (Table 9.5). This represents a major step forward in the understanding of observed GMSL change 54 over the 20th century, which is dominated by glacier (52%) and Greenland ice sheet mass loss (29%) and the 55 effect of ocean thermal expansion (32%), with a negative contribution from the land water storage change (- Do Not Cite, Quote or Distribute 9-95 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 14%). While the combined mass loss for Greenland and glaciers is consistent with the SROCC, updates in 2 the underlying datasets lead to differences in partitioning of the mass loss. 3 4 5 9.6.1.2 Global mean sea-level change budget in the satellite era 6 7 The SROCC (Oppenheimer et al., 2019) concluded that GMSL increased at a rate of 3.16 [2.79 to 3.53 very 8 likely range] mm yr-1 in the period 1993 to 2015 (the satellite altimetry era) and a rate of 3.58 [3.10 to 4.06 9 very likely range] mm yr-1 in the period 2006 to 2015 (the GRACE/Argo era) (high confidence). An updated 10 assessment for the periods 1993 to 2018 and 2006 to 2018 yields values of 3.25 [2.88 to 3.61] and 3.69 [3.21 11 to 4.17] mm yr-1 (high confidence) (Table 9.5), with the slightly larger central estimates consistent with the 12 observed acceleration in GMSL rise since the late 1960s (Dangendorf et al., 2019), given the longer 13 assessment periods. Based on the GMSL assessed time series presented in section 2.3.3.3, GMSL 14 acceleration is estimated as 0.075 [0.066 to 0.080] mm yr –2 for 1971 to 2018 and 0.094 [0.082–0.115] mm yr –2 15 for 1993–2018 (high confidence). For the common period of 1993-2010, the assessed rate of GMSL rise 16 based on tide gauge reconstructions (3.19 [1.18 to 5.20] mm yr-1) is consistent with the assessment based on 17 satellite altimetry (2.77 [2.26 to 3.28] mm yr-1), within the estimated uncertainties. 18 19 Since the SROCC, two new estimates of the LWS contribution have been published (Cáceres et al., 2020; 20 Frederikse et al., 2020b) (see Section 9.6.1.1). For the early 21st century (the periods 1993 to 2018 and 2006 21 to 2018) both publications find a positive LWS contribution (Table 9.5), based on the most recent GRACE- 22 derived estimates. This contrasts with the negative LWS contribution presented for the same periods in the 23 SROCC based on (WCRP Global Sea Level Budget Group, 2018b) and reinforces the low confidence 24 assessment of the LWS contribution. 25 26 For both periods in the satellite era, i.e., 1993 to 2018 and 2006 to 2018, the sum of contributions is 27 consistent with the total observed GMSL change (high confidence) (Table 9.5). However, the latter period, 28 which is characterised by improved data quality and coverage associated with satellite and Argo 29 observations, shows much closer agreement in the central estimates. The marginal sea-level budget closure 30 for the period 1993 to 2018 may indicate underestimated uncertainty, which may be structural as well as 31 parametric. The sea-level budget assessments across the various periods in Table 9.5 demonstrate that the 32 acceleration in GMSL rise (Section 2.3.3.3) since the late 1960s is mostly the result of increased ice sheet 33 mass loss. However, all contributions to GMSL rise show their largest rate during 2006 to 2018, with the ice 34 sheets accounting for about 35% of the total change during this period. 35 36 37 [START TABLE 9.5 HERE] 38 39 Table 9.5: Observed contributions to global mean sea level (GMSL) change for five different periods. Values are 40 expressed as the total change (Δ) in the annual mean or year mid-point value over each period (mm) 41 along with the equivalent rate (mm yr-1). The very likely ranges appear in brackets based on the various 42 section assessments as indicated. Uncertainties for the sum of contributions are added in quadrature, 43 assuming independence. Percentages are based on central estimate contributions compared to the central 44 estimate of the sum of contributions. 45 Observed contribution 1901-1990 1971-2018 1993-2018 2006-2018 1901-2018 to GMSL change {9.6.1.1} {CCBox 9.1} {9.6.1.2} {9.6.1.2} {9.6.1.1} Thermal expansion Δ 31.6 47.5 32.7 16.7 63.2 (mm) (Section 2.3.3.1, Table [14.7 to 48.5] [34.3 to 60.7] [23.8 to 41.6] [8.9 to 24.6] [47.0 to 79.4] 2.7) (31.9%) (50.3%) (45.7%) (34.4%) (38.4%) Do Not Cite, Quote or Distribute 9-96 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI mm 0.36 1.01 1.31 1.39 0.54 yr-1 [0.17 to 0.54] [0.73 to 1.29] [0.95 to 1.66] [0.74 to 2.05] [0.40 to 0.68] Glaciers (Excl. peripheral Δ 51.8 20.9 13.8 7.5 67.2 glaciers) (mm) [30.4 to 73.2] [10.0 to 31.7] [10.0 to 17.6] [6.8 to 8.2] [41.8 to 92.6] (Sections 2.3.2.3, 9.5.1.1) (52.3%) (22.1%) (19.3%) (15.4%) (40.8%) mm 0.58 0.44 0.55 0.62 0.57 yr-1 [0.34 to 0.82] [0.21 to 0.67] [0.40 to 0.70] [0.57 to 0.68] [0.36 to 0.79] Greenland ice sheet (Incl. Δ 29.0 11.9 10.9 10.9 40.4 peripheral glaciers) (mm) [16.3 to 41.7] [7.7 to 16.1] [9.0 to 12.8] [9.5 to 12.2] [27.2 to 53.5] (Sections 2.3.2.4.1, 9.4.1.1) (29.3%) (12.6%) (15.2%) (22.3%) (24.5%) mm 0.33 0.25 0.44 0.91 0.35 yr-1 [0.18 to 0.47] [0.16 to 0.34] [0.36 to 0.51] [0.79 to 1.02] [0.23 to 0.46] Antarctic ice sheet (Incl. Δ 0.4 6.8 6.4 6.4 6.9 peripheral glaciers) (mm) [ -8.8 to 9.6] [-3.9 to 17.5 [4.3 to 8.5] [4.8 to 8.0] [-3.8 to 17.5] (Sections 2.3.2.4.2, 9.4.2.1) (0.4%) (7.2%)] (9.0%) (13.1%) (4.2%) mm 0.00 0.14 0.26 0.53 0.06 yr-1 [-0.10 to 0.11] [-0.08 to 0.37] [0.17 to 0.34] [0.40 to 0.66] [-0.03 to 0.15] Land water storage* Δ -13.8 7.3 7.8 7.2 -12.9 (mm) (Section 9.6.1.1) [-31.4 to 3.8] [-2.4 to 16.9] [3.3 to 12.2] [3.8 to 10.6] [-45.8 to 20.0] (-13.9%) (7.7%) (10.8%) (14.8%) (-7.8%) mm -0.15 0.15 0.31 0.60 -0.11 yr-1 [-0.35 to 0.04] [-0.05 to 0.36] [0.13 to 0.49] [0.32 to 0.88] [-0.39 to 0.17] Do Not Cite, Quote or Distribute 9-97 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI Sum of observed Δ 99.0 94.4 71.6 48.7 164.8 contributions (mm) [63.0 to 135.1] [71.7 to 117.1] [60.5 to 82.6] [39.9 to 57.5] [117.0 to 212.5] mm 1.11 2.01 2.86 4.06 1.41 yr-1 [0.71 to 1.52] [1.52 to 2.49] [2.42 to 3.30] [3.32 to 4.79] [1.00 to 1.82] Observed GMSL Δ 120.1T 109.6T&A 81.2A 44.3A 201.9T&A change (mm) [69.3 to 170.8] [72.8 to 146.4] [72.1 to 90.2] [38.6 to 50.0] [150.3 to (Section 2.3.3.3) 253.5] mm 1.35T 2.33T&A 3.25A 3.69A 1.73T&A yr-1 [0.78 to 1.92] [1.55 to 3.12] [2.88 to 3.61] [3.21 to 4.17] [1.28 to 2.17] 1 T, A 2 and T&A indicate assessments based on tide gauge reconstructions (T), satellite altimetry (A), or a 3 combination of both (T&A). The assessment uses tide gauge reconstructions before 1993 and satellite 4 altimetry after 1993. 5 *For the periods 1971-2018, 1993-2018, 2006-2018 and 1901-2018 the Caceres et al. (2020) 6 linear trends are based on the period up to 2016. 7 8 [END TABLE 9.5 HERE] 9 10 11 9.6.1.3 Regional sea-level change in the satellite era 12 13 Regional sea-level changes are resolved by both tide gauge and satellite altimetry observations (Hamlington 14 et al., 2020a). Altimeters have the advantage of quasi-global coverage but are limited to a period (1993- 15 present) in which the forced trend response is just emerging on regional scales (Section 9.6.1.4). An analysis 16 of the local altimetry error budget to estimate 90% confidence intervals on regional sea-level trends and 17 accelerations reports that 98% of the ocean surface has experienced significant sea-level rise over the 18 satellite era (Prandi et al., 2021). The same study finds that sea-level accelerations display a less uniform 19 pattern, with an east/west dipole in the Pacific, a north/south dipole in the Southern Ocean and in the North 20 Atlantic, and 85% of the ocean surface experiencing significant sea-level acceleration or deceleration, above 21 instrumental and post processing noise. Longer records are available from tide gauges, albeit with variable 22 coverage by basin. Regional departures from GMSL rise are primarily driven by ocean transport divergences 23 that result from wind stress anomalies and spatial variability in atmospheric heat and freshwater fluxes 24 (Section 9.2.4). 25 26 The SROCC (Oppenheimer et al., 2019) noted the occurrence of large multiannual sea-level variations in the 27 Pacific, associated with the Pacific Decadal Oscillation (PDO) in particular, and involving the El Niño 28 Southern Oscillation (ENSO), North Pacific Gyre Oscillation (NPGO) and Indian Ocean Dipole (IOD) 29 (Annex IV) (Royston et al., 2018; Hamlington et al., 2020b). There was intensified sea-level rise during the 30 1990s and 2000s, with 10-year trends exceeding 20 mm yr-1 in the western tropical Pacific Ocean, while sea- 31 level trends were negative on the North American west coast. During the 2010s, the situation reversed, with 32 western Pacific sea level falling at more than 10 mm yr-1 (Hamlington et al., 2020b). For the Atlantic Ocean, Do Not Cite, Quote or Distribute 9-98 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 the SROCC described regional sea-level variability as being driven primarily by wind and heat flux 2 variations associated with the North Atlantic Oscillation (NAO) and heat transport changes associated with 3 AMOC variability. During periods of subpolar North Atlantic warming, winds along the European coast are 4 predominantly from the south and may communicate steric anomalies onto the continental shelf, driving 5 regional sea-level rise, with the reverse during periods of cooling (Chafik et al., 2019). High rates of sea- 6 level rise in the North Indian Ocean are accompanied by a weakening summer South Asian monsoon 7 circulation (Swapna et al., 2017). 8 9 The Arctic ocean is typically excluded from global sea-level studies, owing to the uncertainties associated 10 with resolving sea-level in ice-covered regions, strong variations in GRD effects, and uncertain GIA 11 estimates (Box 9.1). Spanning 1991-2018, a very likely sea-level riseof 1.16-1.81 mm yr-1 is observed (Rose 12 et al., 2019). Since the SROCC, the forced response in regional sea level varies in time with the relative 13 influence of different forcing agents (Fasullo et al., 2020). 14 15 The SROCC estimated regional sea-level changes from combinations of the various contributions to sea- 16 level change from CMIP5 climate model outputs, allowing comparison with satellite altimeter and tide- 17 gauge observations. Closure of the regional sea-level budget is complicated by the fact that regional sea-level 18 variability is larger than GMSL variability and there are more processes that need to be considered, such as 19 vertical land movement and ocean dynamical changes (Box 9.1). A number of observation-based studies 20 have focussed on specific areas, such as the Mediterranean (García et al., 2006), the South China Sea (Feng 21 et al., 2012), the US east coast (Frederikse et al., 2017; Piecuch et al., 2018), the North Atlantic basin 22 (Kleinherenbrink et al., 2016) and the Northwestern European continental shelf seas (Frederikse et al., 2016). 23 Studies using tide gauge data and observation-based estimates of the contributions find that, while local 24 agreement is not yet possible, the observational sea-level budget can be closed on a basin scale (Slangen et 25 al., 2014a; Frederikse et al., 2016, 2018, 2020b). A budget analysis for the GRACE era found that the budget 26 closes in some but not all coastal regions: substantial parts of the sea-level change signal in the North 27 Atlantic could not be explained by steric or barystatic changes (Rietbroek et al., 2016). This is in agreement 28 with other work comparing climate model estimates to 20th century tide gauge observations (Meyssignac et 29 al., 2017), where the majority of local spatial variability is determined by the ocean dynamic component. 30 Vertical land movement is another major cause of local spatial variability in sea-level change, and for 31 instance relevant for oceanic islands (Forbes et al., 2013; Martínez-Asensio et al., 2019). In summary, the 32 regional sea-level budget, using either observations or models, can currently only be closed on basin scales 33 (medium confidence), with large uncertainties remaining on smaller scales. 34 35 36 9.6.1.4 Attribution and time of emergence of regional sea-level change 37 38 The SROCC (Oppenheimer et al., 2019) attributed anthropogenic forcing to be the dominant cause of GMSL 39 rise since 1970 (see also Section 3.5.3.2), but detection and attribution (Cross-Working Group Box: 40 Attribution in Chapter 1) of 20th century externally forced regional sea-level changes is more challenging, as 41 regional variability is larger (Section 9.6.1.3), and therefore the signal-to-noise ratio is smaller (Richter and 42 Marzeion, 2014; Monselesan et al., 2015; Palanisamy et al., 2015). Whereas SROCC assessed with high 43 confidence that GMSL rise is attributable to anthropogenic greenhouse gas emissions, they assessed with 44 medium confidence that the regional anomalies in ocean basins are a combination of the response to 45 anthropogenic GHG emissions and internal variability. 46 47 The simulated ocean dynamic and thermosteric response to external forcings during 1861-2005 is larger than 48 simulated internal variability only in the Southern Ocean and North Pacific on a 1° grid (Slangen et al., 49 2015), but on spatial scales exceeding 2000 km a detectable signal is revealed in the last 45 years in 63% of 50 the global ocean area (Richter et al., 2017). The thermosteric change in the upper 700 m in the period 1970- 51 2005 shows similar observed and simulated forced geographical patterns, and anthropogenic forcing 52 accounts for part (North Atlantic, 65%) or all (tropical Pacific, Southern Ocean) of the observed regional- 53 mean (Marcos and Amores, 2014). The influences of greenhouse gases and anthropogenic aerosols can be 54 partially distinguished by considering geographical or vertical ocean temperature variations (Slangen et al., 55 2015; Bilbao et al., 2019; Fasullo et al., 2020). Zonal-mean forced ocean dynamic sea-level change alone is Do Not Cite, Quote or Distribute 9-99 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 not detectable but, using spatial correlation, the global geographical pattern during the altimeter period is 2 detectable in sea-level trends (Fasullo and Nerem, 2018). This pattern may already or will soon be detectable 3 in individual years, based on an analysis of CMIP5 climate model simulations (Bilbao et al., 2015). 4 Anthropogenic forcing, dominated by greenhouse gases, has strengthened the meridional sea-level gradient 5 in the Southern Ocean since the 1960s (Slangen et al., 2015; Bilbao et al., 2019; Fasullo et al., 2020). New 6 evidence finds that observed zonal-mean total sea-level trends during 1993 to 2018 in all basins are 7 inconsistent with unforced variability alone and consistent with the modelled response to external forcing 8 (Richter et al., 2020). 9 10 A region that has been studied intensely in the context of sea-level detection and attribution is the tropical 11 Pacific. Observed sea-level trends in the tropical Pacific show a PDO-like (Annex IV) East-West dipole 12 (with a greater rate of rise in the west, see section 9.6.1.3). This dipole does not occur in CMIP5 simulations 13 with the magnitude and duration that was observed in the 1990s and 2000s, neither in response to historical 14 forcing, nor as internal variability after removing the variability associated with the PDO (Bilbao et al., 15 2015). (Hamlington et al., 2014) did obtain a residual trend pattern for 1993-2010 in the tropical Pacific that 16 may link to anthropogenic warming of the tropical Indian Ocean. Allowing for PDO and ENSO variations, 17 (Royston et al., 2018) describe patches of the Pacific Ocean where the sea-level trend for 1993-2015 is 18 distinguishable from temporally correlated noise. The acceleration in eastern Pacific sea-level rise is largely 19 accounted for by variations resembling PDO and ENSO (Hamlington et al., 2020a). 20 21 In the future, the anthropogenic signal in regional sea-level change from ocean density and dynamics is 22 projected to emerge first in regions with relatively small internal variability, such as the tropical Atlantic 23 Ocean and the tropical Indian Ocean (Jordà, 2014; Lyu et al., 2014; Richter and Marzeion, 2014; Bilbao et 24 al., 2015). The signal is projected to emerge over 50% of the ocean area by the 2040s (Lyu et al., 2014), but 25 in regions where variability is large and projected changes are small, such as the Southern Ocean, the signal 26 will not emerge before late in the century. Adding the projected sea-level change from land ice mass loss and 27 groundwater extraction strengthens and modifies the forced signal, leading to times of emergence 10-20 28 years earlier in most parts of the ocean, except in regions close to sources of mass loss, with emergence over 29 50% of the ocean area by 2020 and nearly everywhere by 2100 (Lyu et al., 2014; Richter et al., 2017) 30 (medium confidence). 31 32 In summary, detection of forced regional changes for some ocean areas in recent decades is possible 33 (medium confidence), but attribution of regional sea-level change to forcings over longer periods (20th 34 century) and for all ocean basins isnot yet possible. 35 36 37 [START CROSS-CHAPTER BOX 9.1 HERE] 38 39 Cross-Chapter Box 9.1: Global energy inventory and sea level budget 40 41 Coordinators: Matthew D. Palmer (UK), Aimée B.A. Slangen (The Netherlands), 42 43 Contributors: Guðfinna Aðalgeirsdóttir (Iceland), Fábio Boeira Dias (Finland/Brazil), Catia M. Domingues 44 (Australia/Brazil), Gerhard Krinner (France), Johannes Quaas (Germany), Lucas Ruiz (Argentina) 45 46 Increased atmospheric greenhouse gas emissions since the 19th century have led to a net positive radiative 47 forcing of Earth’s climate (Sections 2.2, 7.3) and a corresponding accumulation of energy in the Earth 48 System. Quantification of this energy gain is essential to our understanding of observed climate change and 49 for estimates of climate sensitivity (Section 7.5). The global energy inventory is closely linked to our 50 understanding of observed global sea-level change, through the energy associated with loss of land-based ice 51 and the effect of thermal expansion associated with ocean warming (Box 9.1, Sections 2.3.3.1, 9.6.1; Table 52 9.5). 53 54 The Earth System gained substantial energy over the period 1971-2018 (high confidence), with an assessed 55 very likely range of 325 to 546 ZJ or 0.43 to 0.72 W m-2 expressed per unit area of the Earth’s surface Do Not Cite, Quote or Distribute 9-100 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 (Cross-Chapter Box 9.1, Figure 1a; Section 7.2, Box 7.2). Ocean warming dominates the energy inventory 2 change (high confidence), accounting for 91% of the observed energy increase for the period 1971-2018, 3 with upper ocean warming (0-700 m) accounting for 56% (Section 7.2). Much smaller amounts went into 4 melting of ice (3%) and heating of the land (5%) and atmosphere (1%). Overall, the percentage contributions 5 are similar to those reported in IPCC AR5 for the period 1971-2010 (Rhein et al., 2013). 6 7 The observed global mean sea-level (GMSL) budget is assessed through comparison of the sum of individual 8 components of GMSL change with independent observations of total GMSL change from tide gauge and 9 satellite altimeter observations (Cross-Chapter Box 9.1, Figure 1b; Sections 2.3.3, 9.6.1; Table 9.5). The 10 assessed sum of the observed components indicates that GMSL very likely increased by 72 to 117 mm over 11 the period 1971-2018 (Table 9.5), with the largest contributions from ocean thermal expansion (50%) and 12 melting of ice sheets and glaciers (42%). The assessed total GMSL change (Section 2.3.3) for the period 13 1971-2018 has a very likely range of 73-146 mm and as a result the sea-level budget is closed for this period 14 (Cross-Chapter Box 9.1, Figure 1b, Section 9.6.1, Table 9.5). 15 16 The sea-level budget closure demonstrates improved quantification of the processes of observed GMSL 17 change for this period relative to previous IPCC assessments (Church et al., 2013a; Oppenheimer et al., 18 2019). A related assessment presented in Chapter 7 demonstrates closure of the global energy budget (high 19 confidence) (Box 7.2) and strengthens the confidence in scientific understanding of both of these key aspects 20 of climate change. 21 22 23 [START CROSS-CHAPTER BOX 9.1, FIGURE 1 HERE] 24 25 Cross-Chapter 9.1, Figure 1: Global Energy Inventory and Sea Level Budget. a) Observed changes in the global 26 energy inventory for 1971-2018 (shaded time series) with component contributions as 27 indicated in the figure legend. Earth System Heating for the whole period and 28 associated uncertainty is indicated to the right of the plot (red bar = central estimate; 29 shading = very likely range); b) Observed changes in components of global mean sea- 30 level for 1971-2018 (shaded time series) as indicated in the figure legend. Observed 31 global mean sea-level change from tide gauge reconstructions (1971-1993) and satellite 32 altimeter measurements (1993-2018) is shown for comparison (dashed line) as a 3-year 33 running mean to reduce sampling noise. Closure of the global sea-level budget for the 34 whole period is indicated to the right of the plot (red bar = component sum central 35 estimate; red shading = very likely range; black bar = total sea level central estimate; 36 grey sharing = very likely range). Full details of the datasets and methods used are 37 available in Annex I. Further details on energy and sea-level components are reported in 38 Table 7.1 and Table 9.5. 39 40 [END CROSS-CHAPTER BOX 9.1, FIGURE 1 HERE] 41 42 [END CROSS-CHAPTER BOX 9.1 HERE] 43 44 45 9.6.2 Paleo context of global and regional sea-level change 46 47 As the SROCC (Oppenheimer et al., 2019) noted, paleo-sea level records provide information on past ice- 48 sheet changes, and process-based ice sheet models of past warm periods inform equilibrium responses. 49 However, given uncertainties in paleo-sea level and polar paleoclimate and limited temporal resolution of 50 paleo-sea level records, there is low confidence in the utility of paleo-sea level records for quantitatively 51 informing near-term GMSL change. Nonetheless the paleo record does contextualise sea level and can test 52 projection models (see also FAQ 1.3). Proxy constraints on GMSL and global ice volume are assessed in 53 Sections 2.3.2.4. and 2.3.3.3 (see also FAQ 9.1). This section updates prior assessments of drivers of past 54 GMSL changes and climatically coherent areas of relative sea-level variability. GMSL changes are framed in 55 terms of GMST but noting that amplified high latitude warming is a robust equilibrium response to elevated 56 CO2 (Masson-Delmotte et al., 2013): polar air temperatures during past warm periods were up to twice the Do Not Cite, Quote or Distribute 9-101 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 GMST changes shown in Table 9.6. The SROCC assessment that past multi-metre sea-level changes have 2 resulted from significant ice sheet changes beyond those presently observed is confirmed (very high 3 confidence). 4 5 6 [START TABLE 9.6 HERE] 7 8 Table 9.6: Reference ranges of age, global mean surface temperature, atmospheric carbon dioxide concentration, and 9 global mean sea level for the paleo periods discussed in this chapter. 10 Paleo period Years GMST relative CO2 GMSL to 1850-1900 CCB 2.1 Sections 2.2.3.1, Section 2.3.3.3 Section 2.3.1.1 2.2.3.2 Early Eocene 53 – 49 Ma +10 to +18°C 1150 – 2500 ppm +70 to +76m Climatic Optimum EECO Mid-Pliocene 3.3 – 3.0 Ma +2.5 to +4°C 360 – 420 ppm +5 to +25 m Warm Period (MPWP) Marine Isotope ~424 – 395 ka 0.5 ± 1.6 ºCa 265 – 286 ppm +6 to +13 m Stage (MIS) 11 Last Interglacial ~129 – 116 ka +0.5 to +1.5°C 266 – 282 ppm +5 to +10 m (LIG) Last Glacial 21- 19 ka -5 to -7°C 188 – 194 ppm -125 to -134 m Maximum (LGM) Last Deglacial – 193 –> 271 ppm -120 –> -50 m Transition 18 – 11 ka Early Holocene 11.65 – 6.5 ka – 250 – 268 ppm -50 –> -3.5 m Mid Holocene 6.5 – 5.5 ka +0.2 to +1.0°C 260 - 268 ppm -3.5 to +0.5 m Last Millennium 850 to 1850 -0.14 to +0.24 ºC 278 - 285 ppm -0.05 to +0.03 m CE 11 a Based on one study (Irvalı et al., 2020) relative to ~year 2000 SST values 12 13 [END TABLE 9.6] 14 15 16 Mid-Pliocene Warm Period (MPWP) 17 During the MPWP, GMST was 2.5-4°C warmer than 1850-1900 (medium confidence) and GMSL was 18 between 5 and 25 m higher than today (Table 9.6) (medium confidence) (Section 2.3.3.3). The AR5 19 (Masson-Delmotte et al., 2013) concluded that ice-sheet models consistently produce near-complete 20 deglaciation of the Greenland and West Antarctic Ice Sheets, and multi-meter loss of the East Antarctic Ice 21 Sheet in response to MPWP climate conditions. Studies since the AR5 have yielded a consistent but broader 22 range, due in part to larger ensembles exploring more parameters (DeConto and Pollard, 2016; Yan et al., 23 2016; DeConto et al., 2021). Partly on the basis of these studies the SROCC proposed a ‘plausible’ upper 24 bound on GMSL of 25 m (low confidence) with evidence suggesting an Antarctic contribution of anywhere 25 between 5.4 – 17.8 m. 26 27 The MPWP climate had substantial polar amplification (up to 8 ºC above pre-industrial levels in Arctic 28 Russia, (Fischer et al., 2018); Section 7.4.4.1). Ice sheet model simulations indicate that Northern 29 Hemisphere glaciation was limited to high-elevation regions in eastern and southern Greenland (Figure 9.17) 30 (medium confidence)(De Schepper et al., 2014; Yan et al., 2014; Koenig et al., 2015; Dowsett et al., 2016; Do Not Cite, Quote or Distribute 9-102 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Berends et al., 2019) with Northern Hemisphere glaciation only becoming more widespread from the 2 (cooler) late Pliocene (Bachem et al., 2017; Blake-Mizen et al., 2019; Knutz et al., 2019; Sánchez-Montes et 3 al., 2020). Southern Hemisphere glaciation was characterised by an Antarctic ice sheet reduced in volume 4 from present (Figure 9.18) (Dowsett et al., 2016; Berends et al., 2019; Grant et al., 2019; Miller et al., 2020) 5 (medium confidence) with mountain ice fields in the Andes of South America (De Schepper et al., 2014). Ice 6 sheet models are inconsistent in the magnitude of the sea level contribution from Antarctica (DeConto and 7 Pollard, 2016; Yan et al., 2016; Golledge et al., 2017b; Berends et al., 2019; DeConto et al., 2021) but near- 8 field sedimentological reconstructions support precessionally-modulated and eccentricity-paced multi-metre 9 sea level contributions from the Wilkes Subglacial Basin over 3 – 5 kyr (Patterson et al., 2014; Bertram et 10 al., 2018). In summary, under a past warming level of around 2.5 – 4 ºC, ice sheets in both hemispheres were 11 reduced in extent compared to present (high confidence). Proxy-based evidence (Section 2.3.3.3) combined 12 with numerical modelling indicates that, on millennial timescales, the GMSL contribution arising from ice 13 sheets was >5 m (high confidence) or >10 m (medium confidence) (Moucha and Ruetenik, 2017; Berends et 14 al., 2019; Dumitru et al., 2019) (Figures 9.17, 9.18). 15 16 Marine Isotope Stage 11 (MIS 11) 17 The SROCC (Meredith et al., 2019) noted that Greenland may have been ice-free for extensive periods 18 during Pleistocene interglaciations, implying a high sensitivity of the Greenland ice sheet to warming levels 19 close to present day. The AR5 (Church et al., 2013a) assigned medium confidence to a MIS 11 GMSL of 6– 20 15 m above present, requiring a loss of much of the Greenland and West Antarctic ice sheets, and a possible 21 contribution from East Antarctica. High-resolution multi-proxy sea surface temperature reconstructions and 22 climate model simulations concur that MIS 11 was an extremely long interglacial that exhibited positive 23 annual (0.5 ± 1.6 ºC, (Irvalı et al., 2020)) and summer (2.1–3.4 ºC, (Robinson et al., 2017)) temperature 24 anomalies (de Wet et al., 2016). GMSL was 6-13 m above present (medium confidence, Section 2.3.3.3). The 25 Greenland Ice Sheet lost 4.5–6 m (Reyes et al., 2014)) or ca. 6.1 m (3.9–7 m, 95% confidence) sea-level 26 equivalent by ca. 7 kyr after peak summer warmth (Robinson et al., 2017), with marine-based ice from AIS 27 (Blackburn et al., 2020) contributing 6.4–8.8 m sea-level equivalent at this time (Mas e Braga et al., 2021). 28 Agreement between GMSL and ice-sheet reconstructions gives high confidence in identifying a high 29 sensitivity of both ice sheets to the protracted duration of thermal forcing even at low warming levels (Reyes 30 et al., 2014; Robinson et al., 2017; Irvalı et al., 2020; Mas e Braga et al., 2021). Modelled mean mass loss 31 rates for the Greenland Ice Sheet of 0.4 m kyr-1 during MIS 11 (Robinson et al., 2017) are indistinguishable 32 from recent mass loss rates averaged over 1992–2018 (Section 9.4.1.1). In summary, geological 33 reconstructions and numerical simulations consistently show that past warming levels of <2 ºC (GMST) are 34 sufficient to trigger multi-metre mass loss from both the Greenland and Antarctic ice sheets if maintained for 35 millennia (high confidence), in agreement with the SROCC findings for comparable warming levels during 36 MIS 5e, the Last Interglacial. 37 38 Last Interglacial (LIG) 39 The AR5 found the LIG GMSL was >5 m (very high confidence) but <10 m (high confidence). Their best 40 estimate of 6 m was based on two studies (Kopp et al., 2009; Dutton and Lambeck, 2012). The SROCC 41 concluded that during the Last Interglacial the Greenland contribution to the GMSL highstand of 6–9 m 42 increased gradually whereas the Antarctic contribution occurred early, from ca. 129 ka. Due to widely 43 varying reconstructions from model studies (Greenland) and the paucity of direct evidence of ice sheet 44 change (Antarctic), the magnitude of sea-level contributions from both ice sheets was assigned low 45 confidence. 46 47 Since the AR5, information about the LIG, when GMST was about 0.5-1.5°C above 1850-1900 (medium 48 confidence) (Section 2.3.1.1), has improved. The LIG had higher summer insolation than present and polar 49 amplified sea surface and surface air temperatures that reached >1–4 ºC and >3–11 ºC in the Arctic 50 respectively (Landais et al., 2016; Capron et al., 2017; Fischer et al., 2018). Mean annual and maximum 51 summer ocean temperatures peaked early (129 –125 ka) in the interglacial period, reaching 1.1 ± 0.3 ºC 52 above the modern global mean (Shackleton et al., 2020) with summer anomalies of 2.5–3.5 ºC in the 53 Southern Ocean (Bianchi and Gersonde, 2002) and spatially variable timing (Chadwick et al., 2020). It is 54 virtually certain GMSL was higher than today, likely by 5–10 m (medium confidence) (Section 2.3.3.3). 55 Global mean thermal expansion peaked at about 0.9 ± 0.3 m early in the LIG (~129 ka), declining to modern Do Not Cite, Quote or Distribute 9-103 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 levels by about 127 ka (Shackleton et al., 2020). With no more than 0.3 ± 0.1 m of GMSL rise from glaciers 2 (Section 9.5.1), at most 1.0 ± 0.3 m of the GMSL rise originated from sources other than the polar ice sheets. 3 4 Recent LIG ice sheet simulations agree that peak loss from the Greenland Ice sheet occurred late (125–120 5 ka) (Goelzer et al., 2016; Tabone et al., 2018; Plach et al., 2019) when Northern Hemisphere insolation was 6 greater than at present (Capron et al., 2017) (medium confidence), consistent with inferences from marine 7 sediment records (Hatfield et al., 2016; Irvalı et al., 2020) and far-field GMSL indicators (Rohling et al., 8 2019). Best estimates of the GMSL contribution from Greenland (Figure 9.17) differ between models: <=1 m 9 (Albrecht et al., 2020; Clark et al., 2020), 1–2 m (Calov et al., 2015; Goelzer et al., 2016; Bradley et al., 10 2018), up to 3 m (Tabone et al., 2018; Plach et al., 2019), >5 m (Yau et al., 2016). There is high confidence 11 that the response time of the Greenland Ice Sheet to LIG warming was multi-millennial and high confidence 12 that it contributed to LIG GMSL change, but low agreement in the contribution magnitude. 13 14 Far-field GMSL records suggest that the Antarctic Ice Sheet contributed to LIG sea level from 129.5–125 ka 15 (Figure 9.18) but direct evidence is sparse. Thinning of part of the West Antarctic Ice Sheet is interpreted 16 from a 130 ka–80 ka hiatus in the Patriot Hills horizontal ice core record (Turney et al., 2020). Marine 17 sediment records suggest a dynamic response of the Wilkes Subglacial Basin (WSB) of the East Antarctic Ice 18 Sheet during this period indicating a response timescale of 1000–2500 yr (Wilson et al., 2018), consistent 19 with modelling studies (Mengel and Levermann, 2014; Golledge et al., 2017b; Sutter et al., 2020). Isotopic 20 changes in the Talos Dome ice core are inconsistent with local surface lowering, limiting retreat to 0.4–0.8 m 21 SLE from this sector (Sutter et al., 2020). Ice sheet models forced with unmodified atmosphere–ocean 22 models (Goelzer et al., 2016; Clark et al., 2020) simulate 3–4.4 m sea-level equivalent mass loss, primarily 23 from the West Antarctic Ice Sheet, with no retreat in WSB (e.g., Figure 9.18). Models forced with proxy- 24 based or ad hoc LIG ocean temperature anomalies (DeConto and Pollard, 2016; Sutter et al., 2016) indicate 25 collapse of West Antarctica under 2–3 ºC ocean forcing yielding 3–7.5 m sea-level contribution, but modest 26 or no retreat in the WSB. Based on limited evidence and limited agreement between models, there is low 27 confidence in both the magnitude and timing of LIG mass loss from the Antarctic Ice Sheet. 28 29 In summary, paleo-environmental and modelling studies both indicate that under past warming of the level 30 achieved during the LIG (ca. 0.5–1.5 ºC) it is likely that both the Greenland and Antarctic ice sheets 31 responded dynamically over multiple millennia (high confidence) 32 33 Last Glacial Maximum (LGM) 34 At the LGM geological proxies and GIA models indicate that GMSL was 125-134 m below present (Section 35 2.3.3.3; Figures 9.17, 9.18). New studies have not changed the conclusions of the AR5 regarding the size or 36 timing of the LGM and last glacial termination but have further examined the LGM sea-level budget. Based 37 on a synthesis of multiple prior studies, (Simms et al., 2019) estimated central 67% probability contributions 38 to the LGM lowstand of 76 ± 7 m from the North American Laurentide ice sheet, 18 ± 5 m from the Eurasian 39 ice sheet, 10 ± 2 m from Antarctica, 4 ± 1 m from Greenland, 5.5 ± 0.5 m from glaciers, and 2.4 ± 0.3 m due 40 to an increase in ocean density. Of the residual, up to about 1.4 m may be ascribed to groundwater, leaving a 41 shortfall of 16 ± 10 m yet to be allocated among land ice reservoirs or lakes. 42 43 Last Deglacial Transition: Meltwater pulse 1A (MWP-1A) 44 During MWP-1A, GMSL very likely (medium confidence) rose by 8–15 m (Liu et al., 2016). Consistent with 45 the AR5, the drivers of this rapid rise remain ambiguous. The spatial patterns of relative sea-level change 46 over this interval are inadequately observed to constrain the relative contributions of the North American and 47 Antarctic ice sheets (Liu et al., 2016). Modelling studies of the North American ice sheet permit a 3-6 m 48 (Gregoire et al., 2016) or 6-9 m contribution over the duration of MWP-1A (Tarasov et al., 2012). 49 Sedimentological evidence (Weber et al., 2014; Bart et al., 2018) provides near-field evidence for an 50 Antarctic contribution, consistent with modelling studies (Golledge et al., 2014; Stuhne and Peltier, 2015), 51 but does not constrain the magnitude of the contribution. A recent statistical analysis of Norwegian Sea and 52 Arctic Ocean sediments suggests a 3-7 m contribution from the Eurasian Ice Sheet (Brendryen et al., 2020), a 53 possibility not considered in the AR5 or the meta-analysis of (Liu et al., 2016). In summary, MWP-1A 54 appears to have been driven by a combination of melt in North America (high confidence), Eurasia (low 55 confidence), and Antarctica (low confidence), but the budget is not closed. Do Not Cite, Quote or Distribute 9-104 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Holocene 2 Around half (∼50–60 m) of the GMSL rise since the LGM occurred during the early Holocene at a sustained 3 rate of ∼15 m kyr-1 from ∼11.4–8.2 ka (Lambeck et al., 2014), possibly punctuated by abrupt meltwater 4 pulses (Smith et al., 2011; Carlson and Clark, 2012; Törnqvist and Hijma, 2012; Harrison et al., 2019). An 5 abrupt ~1.1 m sea-level rise at ~8.2 ka was associated with drainage of the pro-glacial lakes Agassiz and 6 Ojibway, attributed to accelerated melt from collapsing Laurentide Ice Sheet ice saddles (Matero et al., 7 2017). The Laurentide Ice Sheet provided the greatest contribution (27 m) to early Holocene GMSL (Peltier 8 et al., 2015; Roy and Peltier, 2017), the Scandinavian ice sheet contributed ~2 m from the beginning of the 9 Holocene until its demise by ~10.5 ka, (Cuzzone et al., 2016), whilst the Barents Sea ice sheet contributed a 10 small but unknown amount (Patton et al., 2015, 2017; Auriac et al., 2016). The Greenland Ice Sheet 11 contributed ~4 m, consistent with ice thinning rates inferred from the Camp Century ice core (Lecavalier et 12 al., 2017; McFarlin et al., 2018). Recent estimates of Antarctic contributions during the early Holocene vary 13 considerably from ~1.2 m to ~8.5 m (Whitehouse et al., 2012; Ivins et al., 2013; Argus et al., 2014; Briggs et 14 al., 2014; Golledge et al., 2014; Pollard et al., 2016; Roy and Peltier, 2017; Albrecht et al., 2020). In 15 summary, the early Holocene was characterised by steadily rising GMSL as global ice sheets continued to 16 retreat from their LGM extents. This steady rise was punctuated by abrupt pulses during episodes of rapid 17 meltwater discharge. 18 19 In the middle Holocene, GMST peaked at 0.2-1.0°C higher than 1850-1900 temperature between 7 and 6 ka 20 (Section 2.3.1.1.2). GMSL rise slowed coincidently with final melting of the Laurentide ice sheet by 6.7 ± 21 0.4 ka (Ullman et al., 2016), after which only Greenland and Antarctic ice sheets could have contributed 22 significantly. At 6 ka, GMSL was -3.5 to +0.5 m (medium confidence) (Section 2.3.3.3). Simulations of the 23 Holocene Thermal Maximum (HTM) give a Greenland ice sheet broadly consistent with geological 24 reconstructions so, despite uncertainties regarding the timing of minimum ice sheet volume and extent, there 25 is medium confidence that minima were reached at different times in different areas during the period 8-3 ka 26 BP (Larsen et al., 2015; Young and Briner, 2015; Briner et al., 2016). Geochronological and numerical 27 modelling studies indicate that it is likely (medium confidence) that the period of smaller-than-present ice 28 extent in all sectors of Greenland persisted for at least 2000 to 3000 years (Larsen et al., 2015; Young and 29 Briner, 2015; Briner et al., 2016; Nielsen et al., 2018). Based on ice sheet modelling and 14C dating 30 (Kingslake et al., 2018) suggested that West Antarctic grounding lines retreated prior to ~10 ka BP, followed 31 by a readvance. Other studies from the same region conclude that retreat was fastest from 9 to 8 ka BP 32 (Spector et al., 2017), or from 7.5 to 4.8 ka BP (Venturelli et al., 2020). Marine geological evidence indicates 33 open marine conditions east of Ross Island by 8.6 ± 0.2 ka BP (McKay et al., 2016). In the western Weddell 34 Sea, (Johnson et al., 2019) reported rapid glacier thinning from 7.5 to 6 ka BP. (Hein et al., 2016) concluded 35 that the fastest thinning further south took place from 6.5 to 3.5 ka BP, potentially contributing 1.4-2 m to 36 GMSL. Geophysical data indicate stabilisation or readvance in this area around 6 ± 2 ka BP (Wearing and 37 Kingslake, 2019). In coastal Dronning Maud Land (East Antarctica) rapid thinning occurred 9 to 5 ka BP 38 (Kawamata et al., 2020), whereas glaciers in the Northern Antarctic Peninsula receded during the period 11 39 to 8 ka BP and readvanced to their maximal extents by 7 to 4 ka BP (Kaplan et al., 2020). In summary, 40 higher-than-pre-industrial GMST during the mid Holocene coincided with recession of the Greenland Ice 41 Sheet to a smaller-than-present extent (high confidence). Multiple lines of evidence give high confidence that 42 thinning or retreat in parts of Antarctica during the Holocene took place at different times in different places, 43 but limited data means there is only low confidence in whether or not the ice sheet as a whole was smaller 44 than present during the mid Holocene. 45 46 In summary, both proxies and model simulations indicate that GMSL changes during the early to mid 47 Holocene were the result of episodic pulses, due to drainage of meltwater lakes, superimposed on a trend of 48 steady rise due to continued ice sheet retreat (high confidence). 49 50 The combination of tide gauge observations and geological reconstructions indicates that a sustained 51 increase of GMSL began between 1820 and 1860 and led to a 20th century GMSL rise that was very likely 52 (high confidence) faster than in any preceding century in the last 3000 yr (Section 2.3.3.3). At a regional 53 level, tide gauge and geological data from the North Atlantic and Australasia show inflections in relative sea- 54 level trends between 1895-1935, with an increase of 0.8 to 2.5 mm yr-1 across the inflection (Gehrels and 55 Woodworth, 2013). A statistical meta-analysis of globally distributed geological and tide gauge data (Kopp Do Not Cite, Quote or Distribute 9-105 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 et al., 2016) found that, in all twenty examined regions with geological records stretching back at least 2000 2 years, the rate of RSL rise in the 20th century was greater than the local average over 0-1700 CE. In four of 3 the twenty regions, all in the North Atlantic (Connecticut, New Jersey, North Carolina, and Iceland), the 19th 4 century rate was also greater than the 0-1700 CE average (90% confidence interval). In summary, rates of 5 RSL rise exceeding the pre-industrial background rate of rise are apparent in parts of the North Atlantic in 6 the nineteenth century (medium confidence) and in most of the world in the twentieth century (high 7 confidence). 8 9 10 9.6.3 Future sea-level changes 11 12 This section first assesses sea-level projections since the AR5 (Church et al., 2013a) (and including the 13 SROCC (Oppenheimer et al., 2019)) based on Representative Concentration Pathways (Section 9.6.3.1). 14 Process-level assessments in sections 9.2.4, 9.4.1.3, 9.4.1.4, 9.4.2.5, 9.4.2.6 and 9.5.1.3 are synthesised 15 (Section 9.6.3.2) to produce new global-mean and regional sea-level projections based on the Shared 16 Socioeconomic Pathways up to 2150 (Section 9.6.3.3) and on global warming levels up to 2100 (Section 17 9.6.3.4). Long-term global mean sea-level projections, both at 2300 and on multimillennial timescales, are 18 also assessed (Section 9.6.3.5). 19 20 In sections 9.6.3.3 and 9.6.3.4, likely ranges of the new global-mean sea-level projections are presented, 21 incorporating only processes in whose projections there is at least medium confidence, consistent with 22 headline projections in the AR5 and the SROCC. As emphasized by the SROCC, there is a substantial 23 likelihood that sea level rise will be outside the likely range. As described in Box 1.1, since the definition of 24 ‘likely’ refers to at least 66% probability, there may be as much as a 34% probability that the processes in 25 which there is at least medium confidence will generate outcomes outside the likely range. Furthermore, 26 additional processes in which there is low confidence (Section 9.4.2.4; Box 9.4) may also contribute to sea- 27 level change. The presentation of likely sea-level change (Tables 9.8-9.9 and in Figures 9.27, 9.29) is 28 therefore accompanied by a low confidence range intended to reflect potential contributions from additional 29 processes under high-emissions scenarios. The low confidence range incorporates ice-sheet projections based 30 on both structured expert judgement (i.e., a formal, calibrated method of combining quantified expert 31 assessments that incorporate all potential processes) and projections from an Antarctic ice sheet model that 32 includes the Marine Ice Cliff Instability (a specific uncertain process not generally included in ice sheet 33 models; Section 9.4.2.4). 34 35 36 9.6.3.1 Global mean sea level projections based on the Representative Concentration Pathways 37 38 The AR5 (Church et al., 2013a) generated global mean sea level (GMSL) projections for the Representative 39 Concentration Pathways (RCPs) by combining information from CMIP5 climate models with glacier and 40 ice-sheet surface mass balance models and assessments of projected ice-sheet dynamic and land-water 41 storage contributions (Section 9.6.3.2). The SROCC (Oppenheimer et al., 2019) updated the AR5 projections 42 based upon a revised assessment of the Antarctic ice sheet contribution to GMSL rise. The AR5 and the 43 SROCC employ a baseline period of 1986 to 2005, which is updated in this report to a baseline period of 44 1995 to 2014 (Section 1.4.1). Between these two periods, GMSL rose by 3 cm and this correction is applied 45 to projections from previous reports to allow comparison (Table 9.8). Accounting for this shift, the 46 conclusion of the SROCC is that, with medium confidence, GMSL will rise between 0.40 m (0.26–0.56 m, 47 likely range) (RCP 2.6) and 0.81 m (0.58–1.07 m, likely range) (RCP 8.5) by 2100 relative to 1995-2014. 48 The AR5 and the SROCC projections of GMSL for the 2007-2018 period have been shown to be consistent 49 with observed trends in GMSL and regional weighted mean tide gauges (Wang et al., 2021a). 50 51 Since the AR5, , a number of projections of GMSL rise have been published based on the RCPs (Slangen et 52 al., 2014a; Kopp et al., 2014, 2017; Grinsted et al., 2015; Jackson et al., 2016; Mengel et al., 2016; Bakker et 53 al., 2017; Bittermann et al., 2017; Nauels et al., 2017; Wong et al., 2017; Le Bars et al., 2017; Nicholls et al., 54 2018; Goodwin et al., 2018; Le Cozannet et al., 2019; Palmer et al., 2020)(see (Garner et al., 2018) for a 55 database; Tables 9.SM.5, 9.SM.6). Some studies also produced associated global sets of regional projections Do Not Cite, Quote or Distribute 9-106 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 (Kopp et al., 2014, 2017; Slangen et al., 2014a; Le Cozannet et al., 2019; Palmer et al., 2020). Since the 2 SROCC, (Le Cozannet et al., 2019) focussed on the low end of the probability distribution of GMSL rise, 3 (Palmer et al., 2020) extended projections beyond 2100 using a climate model emulator (Cross-Chapter Box 4 7.1) and (Horton et al., 2020) conducted a survey of 106 sea-level experts, providing additional context for 5 interpreting sea-level rise projections for 2100 and 2300. 6 7 As noted by the SROCC, the largest differences between projections of GMSL in 2100 are due to the ice 8 sheet projection method, which generally fall into one of three categories: (1) projections from ice sheet 9 models that represent processes in which there is at least medium confidence (Sections 9.4.1.2, 9.4.2.2), (2) 10 projections from an Antarctic ice-sheet model that incorporates the Marine Ice Cliff Instability (MICI) 11 (Section 9.4.2.4) (DeConto and Pollard, 2016), or (3) projections based on structured expert judgement (SEJ) 12 (Sections 9.4.1.3; 9.4.1.4; 9.4.2.5; 9.4.2.6) (Bamber and Aspinall, 2013; Bamber et al., 2019). Low 13 confidence is ascribed to projections incorporating MICI because there is low confidence in the current 14 ability to quantify MICI (Section 9.4.2.4). Low confidence is also ascribed to projections based on SEJ, 15 because individual experts participating in the SEJ study may have incorporated processes in whose 16 quantification there is low confidence, and the experts’ reasoning has not been examined in detail. In general, 17 the range of GMSL projections based upon ice-sheet models not incorporating MICI overlaps with but is 18 lower than projections incorporating MICI or employing SEJ (Figure 9.25). 19 20 There is high agreement across published GMSL projections for 2050 and there is little sensitivity to 21 emissions scenario (Figure 9.25, left panel). Up to 2050, projections are broadly consistent with 22 extrapolation of the observed acceleration of GMSL rise (Sections 2.3.3.3, 9.6.1.1, 9.6.1.2). Considering 23 only projections incorporating ice-sheet processes in whose quantification there is at least medium 24 confidence, the GMSL projections for 2050, across all emissions scenarios, fall between 0.1 and 0.4 m (5th— 25 95th percentile range). Projections incorporating MICI or SEJ do not extend this range under RCP 2.6 or RCP 26 4.5 and extend the upper part of the range to 0.6 m under RCP 8.5. On the basis of these studies, we 27 therefore have high confidence that GMSL in 2050 will be between 0.1 and 0.4 m higher than in 1995 to 28 2014 under low and moderate emissions scenarios and between 0.1 and 0.6 m under high emissions 29 scenarios. 30 31 Conversely, there is low agreement across published GMSL projections for 2100, particularly for higher 32 emissions scenarios, as well as a higher degree of sensitivity to the choice of emissions scenario (Figure 33 9.25, right panel). Considering only projections representing processes in whose quantification there is at 34 least medium confidence, the GMSL projections for 2100 fall between 0.2 and 1.0 m (5th—95th percentile 35 range) under RCP 2.6 and RCP 4.5, and between 0.3 and 1.6 m under RCP 8.5. Considering also projections 36 incorporating MICI or SEJ (low confidence), the projections for 2100 fall between 0.2 and 1.0 m (5th—95th 37 percentile range) under RCP 2.6, 0.2 and 1.6 m under RCP 4.5, and 0.4 and 2.4 m under RCP 8.5. In 38 summary, RCP-based projections published since the AR5 show high agreement for 2050, but exhibit broad 39 ranges and low agreement for 2100, particularly under RCP 8.5. 40 41 42 [START FIGURE 9.25 HERE] 43 44 Figure 9.25: Literature global mean sea level (GMSL) projections (m) for 2050 (left) and 2100 (right) since 1995- 45 2014, for RCP 8.5/SSP5-8.5 (top set), RCP 4.5/SSP2-4.5 (middle set), and RCP 2.6/ SSP1-2.6 46 (bottom set). Projections are standardised to account for minor differences in time periods. Thick bars 47 span from the 17th-83rd percentile projections, and thin bars span the 5th-95th percentile projections. The 48 different assessments of ice sheet contributions are indicated by ‘MED’ (ice sheet projections including 49 only processes in whose quantification there is medium confidence), ‘MICI’ (ice sheet projections which 50 incorporate Marine Ice Cliff Instability), and ‘SEJ’ (structured expert judgement (SEJ) to assess the 51 central range of the ice-sheet projection distributions). ‘Survey’ indicates the results of a 2020 survey of 52 sea-level experts on GMSL rise from all sources (Horton et al., 2020). Projection categories incorporating 53 processes in which there is low confidence (‘MICI’ and ‘SEJ’) are lightly shaded. Dispersion among the 54 different projections represents deep uncertainty, which arises as a result of low agreement regarding 55 appropriate conceptual models describing ice sheet behaviour and low agreement regarding probability 56 distributions used to represent key uncertainties. Individual studies are shown in Tables 9.SM.5, 9.SM.6. Do Not Cite, Quote or Distribute 9-107 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Further details on data sources and processing are available in the chapter data table (Table 9.SM.9). 2 3 [END FIGURE 9.25 HERE] 4 5 6 9.6.3.2 Drivers of projected sea-level change 7 8 This section describes the choices made for the contributions to the updated global mean and regional sea- 9 level projections (Section 9.6.3.3) based on assessments in this report and compares the updated projections 10 to the AR5 (Church et al., 2013a) and the SROCC (Oppenheimer et al., 2019) (Tables 9.7, 9.8). Since there 11 is no single model that can directly compute all of the contributions to sea-level change (Box 9.1), the 12 contributions to sea level are computed separately and then combined (Tables 9.8, 9.9). For consistency with 13 global surface air temperature (GSAT) projections (Section 4.3.1.1) and assessment of equilibrium climate 14 sensitivity (ECS) and transient climate response (TCR) (Section 7.5), temperature-dependent projections 15 (thermal expansion, ice sheets, glaciers) are forced by GSAT projections from a two-layer energy budget 16 emulator (Smith et al., 2018) that is calibrated to be consistent with the assessment of ECS and TCR (Box 17 7.1, Supplementary Material 7.SM.2). Throughout, likely ranges are assessed based upon the combination of 18 uncertainty in the GSAT distribution and uncertainty in the relationships between GSAT and changes to 19 individual components. In general, 17th-83rd percentile results, incorporating both GSAT and sea-level 20 process uncertainty, are interpreted as likely ranges. This is distinct from the approach used by the AR5, 21 which interpreted the 5th-95th percentile range of CMIP5 projections and therefore of GMSL projections 22 driven by them as likely ranges. The shift in interpretation here is consistent with the use of the emulator for 23 GSAT (Box 4.1, Cross-Chapter Box 7.1). Very likely ranges are not assessed because of the potential for 24 processes in whose projections there is currently low confidence to substantially augment total projected 25 GMSL change. 26 27 28 [START TABLE 9.7 HERE] 29 30 Table 9.7: Methods used to project the drivers of GMSL and RSL change in the SSP- and warming-level-based 31 projections of GMSL, RSL and ESL change. Section numbers indicate location of primary assessment 32 text. 33 Driver of Global- SROCC Projection Method AR6 Projection method Mean or Regional Sea-Level change Thermal expansion CMIP5 ensemble Two-layer emulator with climate sensitivity (Section 9.2.4.1) drift-corrected zostoga, with calibrated to the AR6 assessment surrogates derived from (Supplementary Material 7.SM.2) and expansion climate system heat content coefficients calibrated to emulate CMIP6 models where not available (Supplementary Material 9.SM.4.2, 9.SM.4.3) Greenland ice sheet Surface mass balance: Medium confidence processes up to 2100: (excluding scaled cubic polynomial fit to Emulated ISMIP6 simulations (Box 9.3) peripheral glaciers) GMST (Edwards et al., 2021) (Section 9.4.1.3; 9.4.1.4) Dynamics: Medium confidence processes after 2100: Quadratic function of time, Parametric model fit to ISMIP6 simulations up calibrated based on to 2100, extrapolated based on either constant multimodel assessment post-2100 rates or a quadratic interpolation to the multimodel assessed 2300 range (Supplementary Material 9.SM.4.4) Low confidence processes: Do Not Cite, Quote or Distribute 9-108 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI Structured expert judgement (Bamber et al., 2019) Antarctic ice sheet Multimodel assessment Medium confidence processes up to 2100: p-box (excluding including (1) Emulated ISMIP6 simulations peripheral (Edwards et al., 2021) and (2) LARMIP-2 glaciersa) simulations (Levermann et al., 2020a) (Section 9.4.2.5; augmented by AR5 surface mass balance model 9.4.2.6) (Box 9.3) Medium confidence processes after 2100: p-box including (1) AR5 parametric AIS model and (2) LARMIP-2 simulations augmented by AR5 surface mass balance model, with both methods extrapolated based on either constant post-2100 rates or a quadratic interpolation to the multimodel assessed 2300 range (Section 9.6.3.2) Low confidence processes: (1) Single ice-sheet-model ensemble simulations incorporating Marine Ice Cliff Instability (DeConto et al., 2021) and (2) structured expert judgement (Bamber et al., 2019) Glaciers (including Power law function of Up to 2100: peripheral glaciers) integrated GMST fit to Emulated GlacierMIP (Marzeion et al., 2020; (Section 9.5.1.3) glacier models Edwards et al., 2021) simulations (Box 9.3) Beyond 2100: AR5 parametric model re-fit to GlacierMIP (Marzeion et al., 2020) (Supplementary Material 9.SM.4.5) Land water storage Groundwater depletion: Groundwater depletion: (Section 9.6.3.2) combination of (1) Population/groundwater depletion relationship continuation of early 21st calibrated based on (Konikow, 2011; Wada et century trends and (2) land- al., 2012, 2016) surface hydrology models (Wada et al., 2012) Water impoundment: Population/dam impoundment relationship Water impoundment: calibrated based on (Chao et al., 2008), adjusted combination of (1) for new construction following (Hawley et al., continuation of historical rate 2020) for 2020 to 2040 and (2) assumption of no net impoundment after 2010 Ocean dynamic sea CMIP5 ensemble zos field Distribution derived from CMIP6 ensemble zos level after polynomial drift field after linear drift removal (Supplementary (Section 9.2.4.2) removal Material 9.SM.4.2, 9.SM.4.3) Gravitational, Sea-level equation solver (Slangen et al., 2014a) driven by projections of ice rotational, and sheet, glacier, and land water storage changes deformational effects (Section 9.6.3.2) Do Not Cite, Quote or Distribute 9-109 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI GIA model, with ice history Spatiotemporal statistical model of tide-gauge Glacial isostatic from mean of ANU and ICE- data (updated from (Kopp et al., 2014)) adjustment and 5G (Supplementary Material 9.SM.4.6) other drivers of vertical land motion (Section 9.6.3.2) 1 2 aIce sheet models include some of the larger islands in the Antarctic periphery, so there is some overlap in the projected glacier 3 contribution and the projected Antarctic contribution, but the effect of this is estimated to be of order 0.5-1 cm or less (Edwards et 4 al., 2021). 5 6 [END TABLE 9.7 HERE] 7 8 9 Global Mean Thermosteric Sea-Level Rise 10 In the AR5 and the SROCC, global mean thermosteric sea-level rise was derived from the 21 members of the 11 CMIP5 ensemble that provided the required variables (Section 9.2.4.1). The AR5 and the SROCC removed 12 drift estimated based on a pointwise polynomial fit to pre-industrial control simulations. They extended 13 projections to scenarios not provided by the models by calculating the heat content of the climate system 14 from global mean surface temperature and net radiative flux, and converting this to global mean thermosteric 15 sea-level rise using each model’s diagnosed expansion efficiency coefficient. The AR5 and the SROCC 16 derived the associated uncertainties by assuming a normal distribution, with the 5th-95th percentile CMIP5 17 ensemble interpreted as the likely range. In this report, global mean thermosteric sea-level rise is derived 18 from a two-layer energy budget emulator consistent with the assessment of ECS and TCR (Section 9.2.4.1; 19 Supplementary Material 9.SM.4.2, 9.SM.4.3). Despite the change in methodology, this leads to a likely 20 global mean thermosteric contribution (17th-83rd percentile) between 1995 to 2014 and 2100 that represents a 21 minimal change from the AR5 and the SROCC (Table 9.8). 22 23 Greenland ice sheet 24 The AR5 and the SROCC projected the Greenland surface-mass balance using a cubic polynomial fit to a 25 regional climate model as a function of global mean surface temperature (with a log-normal scaling factor 26 reflecting uncertainty in surface-mass balance models, and another scaling factor reflecting the positive 27 feedback of ice-sheet elevation changes on mass loss), and the dynamic contribution was estimated based on 28 a multi-model assessment interpolated as a quadratic function of time. 29 30 For processes in whose projections we have at least medium confidence, the updated projections use 31 emulated ISMIP6 projections of the Greenland ice sheet (Section 9.4.1.3; Box 9.3; Tables 9.2, 9.7; Figure 32 9.17). Since the ISMIP6 emulator does not account for temporal correlation, a parametric fit to the ISMIP6 33 results is employed to calculate rates of change (Supplementary Material 9.SM.4.4). For projections beyond 34 2100 (when the ISMIP6 simulations end), the polynomial fit is extrapolated based on two alternate 35 approaches: (1) an assumption of constant rates of mass change after 2100, and (2) for SSP1-2.6 and SSP5- 36 8.5, a quadratic function of time extending to 2300 based on the multimodel assessment of contributions 37 under RCP 2.6 and RCP 8.5 at 2300 (Section 9.4.1.4). Differences between the two approaches are small up 38 to 2150, and since the latter approach is not available for all scenarios, only the former (constant rates) is 39 used for time-series projections up to 2150. Both approaches are used for examining uncertainty in the 40 timing of different levels of GMSL rise and to inform projections for the year 2300 (Section 9.4.1.4). For 41 2100, the ISMIP6 emulator yields the likely contribution from the Greenland ice sheet shown in Table 9.2 42 and Figure 9.17, representing a slight narrowing from the AR5 projections. 43 44 Antarctic ice sheet 45 For the Antarctic ice sheet, the AR5 applied a temperature-based scaling approach for surface mass balance 46 and a quadratic function of time, calibrated to a multimodel assessment, for dynamic contributions. The Do Not Cite, Quote or Distribute 9-110 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 SROCC used a new assessment based on the results of five process-based studies (Section 9.4.2.5). For 2 processes in whose projections we have at least medium confidence, the likely range projections for the 3 Antarctic ice sheet are based upon 1) the emulated ISMIP6 ensemble and 2) the LARMIP-2 ensemble, 4 augmented with the AR5 parametric Antarctic surface mass balance model. GMSL projections are produced 5 with both distributions and combined in a ‘p-box’ (Kriegler and Held, 2005; Le Cozannet et al., 2017), 6 which represents the upper and lower bounds of the distribution (Section 9.4.2.5, Box 9.3, Table 9.3). A 7 likely range is then identified, spanning the lower of the two 17th percentile projections and the higher of the 8 two 83rd percentile projections5, with the median taken as the mean of the medians of the two projections. 9 Since the ISMIP6 emulator does not account for temporal correlation, the AR5 parametric AIS model is 10 substituted for the emulator in the p-box for rates of change. As the AR5 projections are modestly lower than 11 those from the ISMIP6 emulator, this substitution modestly broadens the likely range at the low end for 12 projections of rate and changes beyond 2100. For projections beyond 2100 (when the ISMIP6 and LARMIP- 13 2 simulations end), the AIS simulations are extrapolated using the same two approaches as the GrIS 14 projections (Section 9.2.4.6). The likely ranges to 2100 are consistent with the SROCC (Table 9.8). 15 16 Low confidence ice sheet projections 17 To test the possible effect of additional ice-sheet processes for which there is low confidence (Sections 18 9.4.1.3, 9.4.1.4, 9.4.2.5, 9.4.2.6, 9.6.3.1; Box 9.4), two additional approaches are considered. For both the 19 Greenland and Antarctic ice sheets, we produce sensitivity cases employing the SEJ projections of (Bamber 20 et al., 2019), mapping 2°C and 5°C stabilization scenarios to SSP1-2.6 and SSP5-8.5, respectively. For the 21 Antarctic ice sheet, we produce an additional sensitivity case using projections, which incorporate the 22 Marine Ice Cliff Instability (DeConto et al., 2021), mapping projections for RCP 2.6 and RCP 8.5 to SSP1- 23 2.6 and SSP5-8.5. For the Greenland ice sheet, the SEJ projections indicate the potential for outcomes 24 outside the corresponding likely ranges (Table 9.8). For the Antarctic ice sheet, there is no evidence from 25 these studies to suggest an important role under lower emissions scenarios for processes in whose projections 26 we have low confidence. By contrast, for SSP5-8.5, the SEJ and MICI projections exhibit 17th-83rd percentile 27 ranges of 0.02-0.56 m and 0.19-0.53 m by 2100, consistent with one another but considerably broader than 28 the likely contribution for medium confidence processes of 0.03 to 0.34 m. This lower level of agreement for 29 higher emissions scenarios reflects the deep uncertainty in the AIS contribution to GMSL change under 30 higher emissions scenarios (Box 9.4). This deep uncertainty grows after 2100: by 2150, under SSP 5-8.5, 31 medium confidence processes likely lead to a -0.1 to 0.7 m AIS contribution, while SEJ and MICI-based 32 projections indicate 0.0-1.1 m and 1.4-3.7 m, respectively. 33 34 Glaciers 35 In the AR5 and the SROCC, global glacier mass changes were derived from a power law of integrated 36 global-mean surface temperature change fit to results from four different glacier models. The updated 37 projections use emulated GlacierMIP projections (Section 9.5.1.3, Box 9.3). Since the GlacierMIP emulator 38 does not account for temporal correlation and terminates, along with the GlacierMIP simulations, in 2100, 39 we employ a parametric fit to the GlacierMIP simulations, with a functional form similar to that employed 40 by the AR5, to calculate rates of change and extrapolate changes beyond 2100 (up to a maximum potential 41 contribution of 0.32 m) (see Supplementary Material 9.SM.4.5). This approach leads to a median glacier 42 contribution that is a minimal change (Table 9.8) from the AR5 and the SROCC and a modest narrowing of 43 likely ranges (Section 9.5.1.3). For RCP 2.6, the AR5 projected 0.10 (0.04-0.16, likely range) m, compared to 44 0.09 (0.07-0.11) m projected here for SSP1-2.6. For RCP 8.5, the AR5 projected a likely contribution of 0.17 45 (0.09-0.25) m, compared to 0.18 (0.15-0.21) m projected here. 46 47 Land water storage 48 In the AR5 and the SROCC, the groundwater depletion contribution to GMSL rise was based on combining 49 results from two approaches, one assuming a continuation of early 21st century trends (Konikow, 2011) and 50 the other using land-surface hydrology models (Wada et al., 2012). Together, these yielded a range of about 51 0.02-0.09 m of GMSL rise by 2080-2099. The rate of water impoundment in reservoirs was likewise based 5 Note that the use of this approach implies that the likely ranges are likely in the use of the term to mean 66–100% probable; this is distinct from the usage in the SROCC, where likely range was defined to have a precise 66% probability. Do Not Cite, Quote or Distribute 9-111 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 on two approaches, one assuming the continuation of the average rate over 1971-2010 (and thus -0.01 to - 2 0.03 m by 2080-2099) (Chao et al., 2008), and the other assuming no net impoundment after 2010 3 (Lettenmaier and Milly, 2009). Together, these yield a GMSL contribution from groundwater impoundment 4 of -0.03 to 0 m. Combining groundwater depletion and water impoundment led the AR5 and the SROCC to 5 infer a projected range of -0.01 to +0.11 m by 2100. 6 7 In the updated projections, a statistical relationship is applied linking historical and future SSP global 8 population to dam impoundment and groundwater extraction (Rahmstorf et al., 2012; Kopp et al., 2014). The 9 population/groundwater depletion relationship is calibrated based on the same studies used in the AR5 10 (Konikow, 2011; Wada et al., 2012), reduced by ~20% to account for water retained on land (Wada et al., 11 2016). The population/dam impoundment relationship is calibrated based upon (Chao et al., 2008). However, 12 while historically dam impoundment has been declining with population, recent literature shows that planned 13 dam construction considerably exceeds the historical trend (Zarfl et al., 2015; Hawley et al., 2020). Over 14 2020-2040, the impoundment contribution to GMSL rise based upon past trends would be about -0.1 mm yr- 1 15 , compared to about -0.5 mm yr-1if all currently planned dams are built (Hawley et al., 2020) and the 16 statistical projection is therefore augmented by an additional -0.4 to 0.0 mm yr-1over 2020-2040 to account 17 for the possible effects of planned dam construction. As in the AR5 and the SROCC, climatically driven 18 changes to LWS have not been included in published sea-level projections, as they are not well quantified 19 (e.g., (Jensen et al., 2019)) or are considered negligible (e.g., permafrost, Section 9.5.2). This approach 20 yields a likely global-mean land water storage contribution (Figure 9.27, Table 9.8) that is slightly lower and 21 narrower than the AR5 and the SROCC likely ranges. Since the projections are explicitly population driven, 22 these projections also exhibit a weak scenario dependence, with a ~0.01 m higher contribution under SSP3 23 than under other scenarios. 24 25 26 [START TABLE 9.8 HERE] 27 28 Table 9.8: Global mean sea-level projections between 1995-2014 and 2100 for total change and individual 29 contributions, median values, (likely) ranges of the process-based model ensemble, for RCP 2.6 (from the 30 AR5 (Church et al., 2013b) and the SROCC (Oppenheimer et al., 2019)) and SSP1-2.6 (this report), and 31 for RCP 8.5 (from the AR5 (Church et al., 2013b) and the SROCC (Oppenheimer et al., 2019)) and 32 SSP5-8.5 (this report). Values for the AR5 (Church et al., 2013b) and the SROCC (Oppenheimer et al., 33 2019) are adjusted from the 1986-2005 baseline used in past reports. Only the Antarctic contribution 34 changed between the AR5 (Church et al., 2013b) and the SROCC (Oppenheimer et al., 2019). Unshaded 35 cells represent processes in which there is medium confidence; shading indicates the inclusion of 36 processes in which there is low confidence. For the MICI and SEJ-based projections, parenthetical 37 numbers represent the 17-83rd percentile of the associated probability distributions, not assessed likely 38 ranges. 39 RCP 2.6 SSP1-2.6 Medium AR5 SROCC confidence MICI SEJ m rel. to 1995-2014 processes Thermal expansion 0.14 (0.10-0.19) 0.14 (0.11-0.18) (9.2.4.1) 0.13 (0.07- Greenland (9.4.1.3) 0.07 (0.03-0.11) 0.06 (0.01-0.10) 0.30) 0.06 (- 0.04 (0.01- 0.11 (0.03- 0.08 (0.06- 0.09 (-0.01- Antarctica (9.4.2.5) 0.04-0.16) 0.11) 0.27) 0.12) 0.25) Glaciers (9.5.1.3) 0.10 (0.04-0.16) 0.09 (0.07-0.11) Land water storage 0.05 (-0.01-0.11) 0.03 (0.02-0.04) (9.6.3.2) Do Not Cite, Quote or Distribute 9-112 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 0.41 0.40 (0.26- 0.44 (0.33- 0.41 (0.35- 0.53 (0.38- Total (2100) (0.25- 0.56) 0.61) 0.48) 0.80) 0.58) 0.56 (0.40- 0.69 (0.46- 0.74 (0.63- 0.84 (0.56- Total (2150) 0.29-0.63 0.73) 1.00) 0.91) 1.34) GMSL rate, 2080- 4.4 (2.0- 4 (2-6) 5.3 (3.3-8.1) 5.2 (4.4-6.2) 6.0 (2.9-1.1) 2100 (mm yr-1) 6.8) RCP 8.5 SSP5-8.5 Medium AR5 SROCC confidence MICI SEJ m rel. to 1995-2014 processes Thermal expansion 0.31 (0.24-0.38) (9.2.4.1) 0.30 (0.24-0.36) 0.23 (0.10- Greenland (9.4.1.3) 0.14 (0.08-0.27) 0.13 (0.09-0.18) 0.59) 0.04 (- 0.12 (0.03- 0.12 (0.03- 0.34 (0.19- 0.21 (0.02- Antarctica (9.4.2.5) 0.08-0.14) 0.28) 0.34) 0.53) 0.56) Glaciers (9.5.1.3) 0.17 (0.09-0.25) 0.18 (0.15-0.21) Land water storage 0.05 (-0.01-0.11) (9.6.3.2) 0.03 (0.02-0.04) 0.71 0.81 (0.58- Total (2100) (0.49- 0.77 (0.63- 0.99 (0.82- 1.01 (0.70- 1.07) 0.95) 1.02) 1.19) 1.61) 1.27 (0.80- 1.35 (1.02- 3.48 (2.58- 1.80 (1.23- Total (2150) 0.34-1.35 1.79) 1.89) 4.83) 2.93) GMSL rate, 2080- 11.2 (7.5- 12.2 (8.8- 23.2 (17.7- 16.1 (9.8- 15 (10-20) 2100 (mm yr-1) 15.7) 17.7) 30.2) 29.1) 1 2 [END TABLE 9.8 HERE] 3 4 5 Ocean dynamic sea level 6 In the AR5 and the SROCC, the ocean dynamic sea-level contribution to relative sea-level (RSL) projections 7 was derived from the CMIP5 ensemble, after removing drift estimated based on pre-industrial control 8 simulations. This report uses updated simulations from the CMIP6 ensemble (Section 9.2.4.2; 9 Supplementary Material 9.SM.4.2) to project the ocean dynamic sea-level contribution to relative sea-level 10 change (Figure 9.26; Section 9.2.4.2). To produce ocean dynamic sea-level projections consistent with the 11 global mean thermosteric projections from the two-layer energy budget emulator, we follow the approach of 12 (Kopp et al., 2014), employing a correlation between global-mean thermosteric sea-level change and ocean 13 dynamic sea level derived from the CMIP6 ensemble (Supplementary Material 9.SM.4.3). Since CMIP6 14 models are of fairly coarse (typically ~100km) resolution, and even the models participating in HighResMIP 15 (near 10km resolution) do not capture all the phenomena that contribute to coastal ocean dynamic sea-level 16 change, there is low confidence in the details of ocean dynamic sea-level change along the coast (Section 17 9.2.3.6) and in semi-enclosed basins, like the Mediterranean, where coarse models can misrepresent key 18 dynamic processes. Regional high-resolution models can improve projections of coastal ocean dynamic sea- 19 level change (Section 12.4) (Hermans et al., 2020), but have not been implemented at a global scale. 20 21 Gravitational, rotational, and deformational (GRD) effects Do Not Cite, Quote or Distribute 9-113 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 GRD effects (Box 9.1) lead to distinct variations in the RSL change pattern, which are similar across a range 2 of benchmarked GRD solvers (Martinec et al., 2018; Palmer et al., 2020). There is high confidence in the 3 understanding of GRD processes. RSL rise associated with GRD is very likely to be largest in the Pacific, 4 due to the combined effects of projected GrIS, AIS and glacier mass loss (high confidence) (e.g., Kopp et al., 5 2014; Slangen et al., 2014a; Larour et al., 2017; Mitrovica et al., 2018). The GRD effect associated with 6 mass loss from an ice sheet is sensitive to the spatial distribution of that mass loss. For example, the GRD 7 contribution to RSL rise in Australia will be larger for Antarctic mass loss sourced fromthe Antarctic 8 Peninsula than for Antarctic mass loss sourced fromThwaites Glacier. In parts of northeastern North 9 America and northwestern Europe, GRD effects associated with mass loss from southern Greenland will lead 10 to a RSL fall, whereas mass loss from northern Greenland will lead to a RSL rise (high confidence) (Larour 11 et al., 2017; Mitrovica et al., 2018) (Figure 9.26). The AR5 and the SROCC computed RSL patterns using a 12 gravitationally self-consistent GRD solver given the amounts, locations and timing of the projected 13 barystatic sea-level changes driven by glaciers, ice sheets and LWS (Church et al., 2013a). A similar GRD 14 solver is used in the updated projections (following (Slangen et al., 2014a)). The Earth model used is based 15 on PREM (Dziewonski and Anderson, 1981), and is elastic, compressible and radially stratified. 16 17 Glacial isostatic adjustment and other drivers of vertical land motion 18 Glacial Isostatic Adjustment (GIA; Box 9.1) leads to vertical land motion (VLM; Box 9.1) and changes in 19 sea-surface height, both of which contribute to RSL change. GIA uncertainty is caused by uncertainty in the 20 rheological structure of the solid Earth, which drives the longer-term viscous Earth deformation, as well as in 21 the modelled global ice history (e.g., Whitehouse, 2018). In the AR5 and the SROCC, GIA contributions to 22 relative sea-level change were calculated using a sea-level equation solver with an ice-sheet history taken as 23 the mean of the ICE5G (Peltier et al., 2015) and ANU (Lambeck et al., 2014) ice sheet models. Since the 24 AR5, new global models are emerging that more rigorously treat ice and Earth structure uncertainty (Caron 25 et al., 2018). However, there is also a growing recognition that lateral variations in Earth structure limit the 26 utilityof global models that treat the solid Earth as though it were laterally uniform (Love et al., 2016; Huang 27 et al., 2019; Li et al., 2020c). 28 29 As noted by the SROCC, VLM from sources other than GIA – including tectonics and mantle dynamic 30 topography, volcanism, compaction, and anthropogenic subsidence – can be locally important, producing 31 VLM rates comparable to or greater than rates of GMSL change. Complete global projections of these 32 processes are not available because of the small spatial scales, the sensitivity of subsidence to local human 33 activities, and the stochasticity of tectonics (Wöppelmann and Marcos, 2016; Oppenheimer et al., 2019). 34 Integrated RSL projections to date have therefore either included only the component of VLM associated 35 with GIA, as in the AR5 and SROCC, or used a constant long-term background rate of change (including 36 both GIA and other long-term drivers of VLM) estimated from historical tide gauge trends (e.g., (Kopp et al., 37 2014)). The updated projections use the second approach and extrapolate the field of long-term background 38 rates of RSL change, including long-term VLM derived from tide gauges, to global coverage using a 39 spatiotemporal statistical approach (Kopp et al., 2014) (Supplementary Material 9.SM.4.6). The combined 40 GIA and long-term VLM is assumed to be constant over the projected period and scenario independent. In 41 areas (e.g., the western Gulf of Mexico) where rapid subsidence occurs in a cluster of tide-gauges, the 42 associated rates are interpolated between the tide gauges. In areas where the available tide-gauges exhibit 43 large, tectonically driven VLM that changes considerably in rate over short distances (e.g., Alaska and the 44 Bering Strait) a sizable uncertainty propagates into the RSL projections (Figure 9.26). Rates of RSL rise are 45 likely to be underestimated due to subsidence in shallow strata that are not recorded by tide gauges (Keogh 46 and Törnqvist, 2019) and in some locations may therefore be minimum values, especially if anomalously 47 high subsidence rates associated with fluid extraction (e.g., (Minderhoud et al., 2017)) are also considered. 48 There is therefore, depending on location, low to medium confidence in the GIA and VLM projections 49 employed in this report. In many regions, higher fidelity projections would require more detailed regional 50 analysis. 51 52 53 54 55 Do Not Cite, Quote or Distribute 9-114 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 [START FIGURE 9.26 HERE] 2 3 Figure 9.26: Median global mean and regional relative sea-level projections (m) by contribution for the SSP1- 4 2.6 and SSP5-8.5 scenarios. (upper time series) Global mean contributions to sea-level change as a 5 function of time, relative to 1995-2014. (lower maps) Regional projections of the sea-level contributions 6 in 2100 relative to 1995-2014 for SSP5-8.5 and SSP1-2.6. Vertical land motion is common to both SSPs. 7 Further details on data sources and processing are available in the chapter data table (Table 9.SM.9). 8 9 [END FIGURE 9.26 HERE] 10 11 12 9.6.3.3 Sea-level projections to 2150 based on SSP scenarios 13 14 Up to 2050, consistent with the AR5 and the SROCC, GMSL projections exhibit little scenario dependence 15 (Figure 9.27, Table 9.9) (high confidence), with likely (medium confidence) sea-level rise between the 16 baseline period (1995 to 2014) and 2050 of 0.19 (0.16-0.25) m under SSP1-2.6 and 0.23 (0.20-0.30) m under 17 SSP5-8.5. These projections fall centrally within the range of published projections for RCP 2.6 and RCP 8.5 18 (Section 9.6.3.1). 19 20 Beyond 2050, the scenarios increasingly diverge. Between the baseline period (1995 to 2014) and 2100, 21 processes in whose projection there is medium confidence drive likely GMSL rise of 0.44 m (0.33-0.61) m 22 and 0.77 (0.63-1.02) m under SSP1-2.6 and SSP5-8.5, respectively (Tables 9.8, 9.9). While derived using 23 substantially updated methods, these projections are broadly consistent with the SROCC, which over this 24 period projected likely GMSL rise of 0.41 (0.26-0.56) m and 0.81 (0.58-1.07) m under RCP 2.6 and RCP 8.5, 25 respectively. They are modestly higher than those of the AR5, which projected likely GMSL rise of 0.41 26 (0.25-0.58) m under RCP 2.6 and 0.71 (0.49-0.95) m under RCP 8.5 (Figure 9.25, Table 9.8). They are also 27 broadly consistent with projections produced by driving the AR5 methods with CMIP6 temperature and 28 thermal expansion projections, which leads to 0.44 (0.27-0.61) m under SSP1-2.6 and 0.73 (0.49-1.02) m 29 under SSP5-8.5 (Hermans et al., 2021). The SSP1-2.6 and SSP5-8.5 projections are consistent with the 30 ranges of published projections for RCP 2.6 and RCP 8.5 that do not incorporate MICI or SEJ (Section 31 9.6.3.1). 32 33 The likely GMSL projections for SSP3-7.0 and SSP5-8.5 are consistent with a continuation of the GMSL 34 satellite-observed rate (very likely 3.25 [2.88-3.61] mm yr-1) and acceleration (very likely 0.094 [0.082-0.115] 35 mm yr-2) of GMSL rise over 1993-2018 (Table 9.5 and Section 2.3.3.3), which would imply a likely GMSL 36 rise of 0.24 m (0.23-0.25 m) by 2050 and 0.73 m (0.69-0.77 m) by 2100. This extrapolation would also 37 imply a likely rate of GMSL rise of 7.5 (7.4-7.6) mm yr-1over 2040-2060 and 11.2 (10.6-11.8) mm yr-1over 38 2080-2100. Over the satellite period, the observed acceleration has been driven primarily by ice-sheet 39 contributions (Section 9.6.1.2; Table 9.5); in the median projections for SSP3-7.0 and SSP5-8.5, these 40 accelerations are projected to continue at a slightly lower level, while the GMSL acceleration is augmented 41 by an acceleration of thermal expansion and glacier loss associated with rising global temperature. Overall, 42 these extrapolations imply that, under SSP1-1.9, SSP1-2.6, and SSP2-4.5, the GMSL acceleration is 43 projected to decrease from its current level. 44 45 46 [START TABLE 9.9 HERE] 47 48 Table 9.9: Global mean sea-level projections for 5 SSP scenarios, relative to a baseline of 1995-2014, in meters. 49 Individual contributions are shown for the year 2100. Median values (likely ranges) are shown. Average 50 rates for total sea-level change are shown in mm yr-1. Unshaded cells represent processes in whose 51 projections there is medium confidence. Shaded cells incorporate a representation of processes in which 52 there is low confidence; in particular, the SSP5-8.5 low confidence column shows the 17th-83rd percentile 53 range from a p-box including SEJ- and MICI-based projections rather than an assessed likely range. 54 Methods are described in 9.6.3.2. 55 Do Not Cite, Quote or Distribute 9-115 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI SSP5-8.5 Low SSP1-1.9 SSP1-2.6 SSP2-4.5 SSP3-7.0 SSP5-8.5 Confidence Thermal 0.12 (0.09- 0.14 (0.11- 0.20 (0.16- 0.25 (0.21- 0.30 (0.24- 0.30 (0.24- expansion 0.15) 0.18) 0.24) 0.30) 0.36) 0.36) 0.05 (0.00- 0.06 (0.01- 0.08 (0.04- 0.11 (0.07- 0.13 (0.09- 0.15 (0.09- Greenland 0.09) 0.10) 0.13) 0.16) 0.18) 0.59) 0.10 (0.03- 0.11 (0.03- 0.11 (0.03- 0.11 (0.03- 0.12 (0.03- 0.19 (0.02- Antarctica 0.25) 0.27) 0.29) 0.32) 0.34) 0.56) 0.08 (0.06- 0.09 (0.07- 0.12 (0.10- 0.16 (0.13- 0.18 (0.15- 0.17 (0.12- Glaciers 0.10) 0.11) 0.15) 0.18) 0.21) 0.22) Land 0.03 (0.02- 0.03 (0.02- 0.03 (0.02- 0.04 (0.02- 0.03 (0.02- 0.03 (0.02- Water 0.04) 0.04) 0.04) 0.05) 0.04) 0.04) Storage Total 0.09 (0.08- 0.09 (0.08- 0.09 (0.08- 0.10 (0.08- 0.10 (0.09- 0.10 (0.09- (2030) 0.12) 0.12) 0.12) 0.12) 0.12) 0.15) Total 0.18 (0.15- 0.19 (0.16- 0.21 (0.18- 0.22 (0.19- 0.23 (0.20- 0.24 (0.20- (2050) 0.23) 0.25) 0.26) 0.28) 0.30) 0.40) Total 0.35 (0.26- 0.39 (0.30- 0.48 (0.38- 0.56 (0.46- 0.64 (0.52- 0.71 (0.52- (2090) 0.49) 0.54) 0.65) 0.74) 0.83) 1.31) Total 0.38 (0.28- 0.44 (0.33- 0.56 (0.44- 0.68 (0.55- 0.77 (0.63- 0.88 (0.63- (2100) 0.55) 0.61) 0.76) 0.90) 1.02) 1.61) Total 0.57 (0.37- 0.69 (0.46- 0.93 (0.67- 1.21 (0.92- 1.35 (1.02- 1.99 (1.02- (2150) 0.85) 1.00) 1.33) 1.67) 1.89) 4.83) Rate (2040- 4.2 (2.9-6.1) 4.9 (3.6-6.9) 5.9 (4.5-8.0) 6.5 (5.1-8.7) 7.3 (5.7-9.8) 7.9 (5.7-16.2) 2060) Rate (2080- 4.3 (2.5-6.6) 5.3 (3.3-8.1) 7.8 (5.3-11.5) 10.4 (7.5-14.9) 12.2 (8.8-17.7) 15.9 (8.8-30.2) 2100) 1 2 [END TABLE 9.9 HERE] 3 4 5 While ice-sheet processes in whose projection there is low confidence have little influence up to 2100 on 6 projections under SSP1-1.9 and SSP1-2.6 (Table 9.9), this is not the case under higher emissions scenarios, 7 where they could lead to GMSL rise well above the likely range. In particular, under SSP5-8.5, low- 8 confidence processes could lead to a total GMSL rise of 0.6-1.6 m over this time period (17th-83rd percentile 9 range of p-box including SEJ- and MICI-based projections), with 5th-95th percentile projections extending to 10 0.5-2.3 m (low confidence). The assessed low confidence range is slightly narrower than but broadly 11 consistent with the full 0.4-2.4 m range of published 5th-95th percentile projections for RCP 8.5 since the 12 AR5 (Section 9.6.3.1), including those based on SEJ or incorporating MICI, and highlights the deep 13 uncertainty in GMSL rise under the highest emissions scenarios (Box 9.4). The assessment of the potential 14 contribution of processes in which there is low confidence to GMSL rise by 2100 is broadly consistent with 15 the assessment of the AR5 (Church et al., 2013a) , which concluded that collapse of marine-based sectors of 16 the Antarctic ice sheet could cause several tenths of a meter of global mean sea-level rise above the likely 17 range. 18 19 While prior assessment reports, starting with the First Assessment Report (Warrick et al., 1990), have 20 focused on projecting GMSL up to the year 2100, time has progressed, and the year 2100 is now within the 21 timeframe of some long-term infrastructure decisions. For this reason, projections up to the year 2150 are 22 also highlighted (Table 9.9). Over this time period, assuming no acceleration in ice-sheet mass fluxes after 23 2100, processes in which there is medium confidence lead to GMSL rise of 0.5-1.0 m under SSP1-2.6 and 24 1.0-1.9 m under SSP5-8.5. Processes in which there is low confidence could drive GMSL rise under SSP5- 25 8.5 to 1.0-4.8 m (17th-83rd percentile) or even 0.9-5.4 m (5th-95th percentile). 26 Do Not Cite, Quote or Distribute 9-116 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 [START FIGURE 9.27 HERE] 2 3 Figure 9.27: Projected global mean sea level rise under different SSP scenarios. Likely global mean sea-level change 4 for SSP scenarios resulting from processes in whose projection there is medium confidence. Projections and 5 likely ranges at 2150 are shown on right. Lightly shaded ranges and thinner lightly shaded ranges on the 6 right show the 17th-83rd and 5th-95th percentile ranges for projections including low confidence processes 7 for SSP1-2.6 and SSP5-8.5 only, derived from a p-box including Structured Expert Judgement and Marine 8 Ice Cliff Instability projections. Black lines show historical GMSL change, and thick solid and dash-dotted 9 black lines show the mean and likely range extrapolating the 1993-2018 satellite altimeter trend and 10 acceleration. Further details on data sources and processing are available in the chapter data table (Table 11 9.SM.9). 12 13 [END FIGURE 9.27 HERE] 14 15 16 Median projected relative sea-level changes are shown in Figure 9.28, with driving factors highlighted in 17 Figure 9.26. Over the 21st century, the majority of coastal locations have a median projected regional sea- 18 level rise within +/- 20% of the median projected GMSL change (medium confidence). Consistent with the 19 AR5, loss of land-ice mass will be an important contributor to spatial patterns in relative sea-level change 20 (high confidence), with ocean dynamic sea-level being particularly important as a dipolar contributor in the 21 Northwest Atlantic, a positive contributor in the Arctic Ocean, and a negative contributor in the Southern 22 Ocean south of the ACC (medium confidence) (Section 9.2.4.2). As today, vertical land motion will remain a 23 major driver of relative sea-level change (high confidence). Uncertainty in relative sea-level projections is 24 greatest in tectonically active areas in which vertical land motion varies over short distances (e.g., Alaska) 25 and in areas potentially subject to large ocean dynamic sea-level change (e.g., the northwestern Atlantic) 26 (high confidence). 27 28 29 [START FIGURE 9.28 HERE] 30 31 Figure 9.28: Regional sea level change at 2100 for different scenarios (with respect to 1995-2014). Median 32 regional relative sea-level change from 1995 to 2014 up to 2100 for (a) SSP1-1.9, (b) SSP1-2.6, (c) SSP2- 33 4.5, (d) SSP3-7.0, (e) SSP5-8.5, and (f) width of the likely range for SSP3-7.0. The high uncertainty in 34 projections around Alaska and the Aleutian Islands arises from the tectonic contribution to vertical land 35 motion, which varies greatly over short distances in this region. Further details on data sources and 36 processing are available in the chapter data table (Table 9.SM.9). 37 38 [END FIGURE 9.28 HERE] 39 40 41 An alternative perspective on uncertainty in future sea-level rise is provided by looking at uncertainty in time 42 rather than elevation; that is, looking at the range of dates when specific thresholds of sea-level rise are 43 projected to be crossed (Figure 9.29). Considering only medium confidence processes, GMSL rise is likely to 44 exceed 0.5 m between about 2080 and 2170 under SSP1-2.6 and between about 2070 and 2090 under SSP5- 45 8.5. It is likely to exceed 1.0 m between about 2150 and some point after 2300 under SSP1-2.6, and between 46 about 2100 and 2150 under SSP5-8.5. It is unlikely to exceed 2.0 m until after 2300 under SSP1-2.6, while it 47 is likely to do so between about 2160 and 2300 under SSP5-8.5. However, processes in whose projections 48 there is low confidence could lead to substantially earlier exceedances under higher emissions scenarios: 49 under SSP5-8.5, 1.0 m could be exceeded by about 2080 and 2.0 m could be exceeded by about 2110 (17th 50 percentile of p-box incorporating projections based on SEJ and MICI), with 5th percentile projections as early 51 as about 2070 for 1.0 m and 2090 for 2.0 m. 52 53 54 [START FIGURE 9.29 HERE] 55 56 Figure 9.29: Timing of when GMSL thresholds of 0.5, 1.0, 1.5 and 2.0 m are exceeded, based upon four different 57 ice-sheet projection methods informing post-2100 projections. Methods are labelled based on their Do Not Cite, Quote or Distribute 9-117 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 treatment of ice sheets. “No acceleration” assumes constant rates of mass change after 2100. “Assessed 2 ice sheet” models post-2100 ice sheet losses using a parametric fit (Supplementary Material 9.SM.4) 3 extending to 2300 based on a multimodel assessment of contributions under RCP2.6 and RCP8.5 at 2300. 4 Structured Expert Judgement (SEJ) employs ice-sheet projections from Bamber et al. (2019) Marine Ice 5 Cliff Instability (MICI) combines the parametric fit (Supplementary Material 9.SM.4) for Greenland with 6 Antarctic projections based on (DeConto et al., 2021). Circles/thick bars/thin bars represent the 50th, 17th- 7 83rd, and 5th-95th percentiles of the exceedance timing for the indicated projection method. Further details 8 on data sources and processing are available in the chapter data table (Table 9.SM.9). 9 10 [END FIGURE 9.29 HERE] 11 12 13 9.6.3.4 Sea-level projections up to 2100 based on global warming levels 14 15 Global warming levels represent a new dimension of integration in the AR6 cycle (Section 1.6.2, Cross- 16 Chapter Box 11.1). The SR1.5 (Hoegh-Guldberg et al., 2018) concluded that, based upon an assessment of 17 GMSL projections published for 1.5°C and 2.0°C scenarios, there is medium agreement that GMSL in 2100 18 would be 0.04 –0.16 m higher in a 2°C warmer world compared to a 1.5°C warmer world based on the 17– 19 83% confidence interval (0.00 – 0.24 m based on the 5–95% confidence interval) with a value of around 0.1 20 m. The SR1.5 did not attempt to standardize the definition of warming-level scenarios, or to examine 21 additional warming levels. No new integrated GMSL projections for 1.5°C or 2.0°C scenarios have been 22 published since the SR1.5. 23 24 Most of the contributors to GMSL are more closely tied to time-integrated GSAT than instantaneous GSAT 25 (Hermans et al., 2021), which means that sea level projections by warming level can only be interpreted if 26 the warming levels are linked to a specific timeframe. Here, the warming level projections are defined based 27 on the 2081-2100 GSAT anomaly (Supplementary Material 9.SM.4.7). Different pathways in GSAT can be 28 followed to reach a certain temperature level, which affects the temporal evolution of the different 29 contributors to sea-level change. For instance, there will be different ice sheet and glacier responses to a fast 30 increase to a peak warming of 2°C in 2050 followed by a plateau or a decrease, compared to a gradual 31 increase to the same level of warming in 2100. The sea-level projections presented might include different 32 pathways to the same warming level in 2100, which is reflected in the uncertainty ranges, and should 33 therefore be interpreted as an illustration of sea-level scenarios under a certain warming level. 34 35 Projections of likely 21st century GMSL rise along climate trajectories leading to different increases in GSAT 36 between 1850-1900 and 2081-2100 are shown in Table 9.10, along with the SSPs for which the temperature- 37 level projections are most closely aligned. For example, considering only processes in which there is medium 38 confidence, from the baseline period (1995 to 2014) up to 2100, GMSL in a 2°C scenario is likely to rise by 39 0.40-0.69, which is intermediate between the projections for SSP1-2.6 and SSP2-4.5. GMSL in a 4°C 40 scenario is likely to rise by 0.58-0.91 m, similar to the projection for SSP3-7.0. Consistent with the 41 discussion in Section 9.6.3.3, there is deep uncertainty in the projections for temperature levels above 3°C, 42 and alternative approaches to projecting ice sheet changes may yield substantially different projections in 43 4°C and 5°C futures. For example, employing SEJ (Bamber et al., 2019) ice-sheet projections instead of the 44 projections for medium confidence processes only leads to a 17th-83rd percentile rise between the baseline 45 period (1995-2014) and 2100 of 0.7-1.6 m rather than 0.7-1.1 m in a 5°C scenario. 46 47 48 [START TABLE 9.10 HERE] 49 50 Table 9.10: Global mean sea-level projections and commitments for exceedance of 5 global warming levels, defined 51 by sorting GSAT change in 2081-2100 w.r.t. 1850-1900. Median values and (likely) ranges are in meters 52 relative to a 1995-2014 baseline. Rates are in mm yr-1. Unshaded cells represent processes in whose 53 projections there is medium confidence. Shaded cells incorporate a representation of processes in which 54 there is low confidence; in particular, the SSP5-8.5 low confidence column shows the 17th-83rd percentile 55 range from a p-box including SEJ- and MICI-based projections rather than an assessed likely range. 56 Methods are described in 9.6.3.2. Do Not Cite, Quote or Distribute 9-118 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 SSP5-8.5 Low 1.5°C 2.0°C 3.0°C 4.0°C 5.0°C Confidence SSP1- SSP2-4.5/SSP3- Closest SSPs SSP1-2.6 SSP3-7.0 SSP5-8.5 2.6/SSP2-4.5 7.0 0.19 (0.16-- 0.20 (0.17-- 0.21 (0.18-- 0.22 (0.19-- 0.25 (0.22-- 0.24 (0.20-- Total (2050) 0.24) 0.26) 0.27) 0.28) 0.31) 0.40) 0.44 (0.34-- 0.51 (0.40-- 0.62 (0.50-- 0.70 (0.58-- 0.81 (0.68-- 0.88 (0.63-- Total (2100) 0.59) 0.69) 0.81) 0.91) 1.05) 1.61) Rate (2040- 4.1 (3.0--5.8) 5.1 (3.8--7.1) 6.0 (4.7--8.2) 6.5 (5.1--8.6) 7.3 (5.8--9.8) 7.9 (5.7--16.2) 2060) Rate (2080- 11.8 (8.6-- 15.9 (8.8-- 4.3 (2.6--6.5) 5.5 (3.5--8.3) 7.9 (5.4--11.6) 9.9 (7.2--14.2) 2100) 17.0) 30.2) 2000-yr 2-3 2-6 4-10 12-16 19-22 commitment 10000-yr 6-7 8-13 10-24 19-33 28-37 commitment 2 3 [END TABLE 9.10 HERE] 4 5 6 9.6.3.5 Multi-century and multi-millennial sea-level rise 7 8 Neither the AR5 nor the SROCC discussed the sea-level commitment associated with historical emissions. 9 Since the AR5, new evidence has suggested that historical emissions up to 2016 will lead to a likely 10 committed sea-level rise (i.e., the rise that would occur in the absence of additional emissions) of 0.7—1.1 m 11 up to 2300, while pledged emissions through 2030 increase the committed rise to 0.8—1.4 m (Nauels et al., 12 2019). 13 14 Between the baseline period (1995 to 2014) and 2300, the AR5 projected a GMSL rise of 0.38-0.82 m under 15 a non-specific low emissions scenario and 0.9-3.6 m under a non-specific high emissions scenario (Table 16 9.11). The SROCC projected 0.6-1.0 m under RCP 2.6 and 2.3-5.3 m under RCP 8.5 (low confidence). RCP- 17 based projections for 2300 published since the AR5 span a broader range, even excluding studies employing 18 SEJ or MICI, with 17-83rd percentile projections ranging from 0.3-2.9 m for RCP 2.6 and 1.7-6.8 m for RCP 19 8.5 (Kopp et al., 2014, 2017, Nauels et al., 2017, 2019; Bamber et al., 2019; Palmer et al., 2020) (Table 20 9.SM.8). Conservatively extending the ISMIP6- and LARMIP-2-based projections beyond 2100 by 21 assuming no subsequent change in ice-sheet mass flux rates (an approach similar to that adopted by (Palmer 22 et al., 2020) for the Greenland Ice Sheet and for the Antarctic Ice Sheet dynamics) leads to a GMSL change 23 up to 2300 of 0.8-2.0 m under SSP1-2.6 and 1.9-4.1 m under SSP5-8.5 (17th-83rd percentile), while 24 incorporating the ice-sheet contributions for 2300 assessed in Section 9.4.1.4 and Section 9.4.2.6 leads to 25 0.6-1.5 m and 2.2-5.9 m, respectively. Incorporating Antarctic results from a model with Marine Ice Cliff 26 Instability (Section 9.4.2.4), using RCP forcing to inform SSP-based projections, leads to 1.4-2.1 m for 27 SSP1-2.6 and 9.5-16.2 m for SSP5-8.5 (DeConto et al., 2021). Incorporating the SEJ-based ice-sheet 28 projections of (Bamber et al., 2019) for 2°C and 5°C stabilization scenarios yields 1.0-3.1 m for SSP1-2.6 29 and 2.4-6.3 m for SSP5-8.5, though because of the differences in scenarios, the SSP1-2.6 estimates may be 30 overestimated and the SSP5-8.5 may be underestimated. The eight-fold uncertainty range across projection 31 methods under SSP5-8.5 reflects deep uncertainty in the multi-century response of ice sheets to strong 32 climate forcing. 33 Do Not Cite, Quote or Distribute 9-119 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Taking into account all these approaches, including published projections for RCP 2.6, under SSP1-2.6 2 GMSL will rise between 0.3 and 3.1 m by 2300 (low confidence). This projection range indicates that, while 3 the SROCC projections under low emissions to 2300 are consistent with no ice sheet acceleration after 2100, 4 there is the possibility of a much broader range of outcomes at the high end, reflected in the range of 5 published GMSL projections. Under SSP5-8.5, GMSL will rise between 1.7 and 6.8 m by 2300 in the 6 absence of MICI and by up to 16 m considering MICI, a wider range than the AR5 or the SROCC 7 assessments but consistent with published projections (low confidence). 8 9 10 [START TABLE 9.11 HERE] 11 12 Table 9.11: Global mean sea-level projections between 1995-2014 and 2300 for total change and individual 13 contributions, for low emissions (from the AR5 (Church et al., 2013b)), RCP 2.6 (from the SROCC 14 (Oppenheimer et al., 2019), and published projections (Table 9.SM.8)) and SSP1-2.6 (this report), and for 15 high emissions (from the AR5(Church et al., 2013b)), RCP 8.5 (from the SROCC (Oppenheimer et al., 16 2019), and published projections (Table 9.SM.8)), and SSP5-8.5 (this report). Values for the AR5 17 (Church et al., 2013b) and the SROCC (Oppenheimer et al., 2019) are adjusted from the 1986-2005 18 baseline used in past reports. Only total values are shown for published ranges. Only the Antarctic 19 contribution changed between the AR5 (Church et al., 2013b) and the SROCC (Oppenheimer et al., 20 2019). If a range is given it is the 17th—83rd percentile range. 21 22 Low RCP 2.6 SSP1-2.6 Assessed No ice-sheet ice-sheet MICI SEJ m rel. to 1995- Post-AR5 acceleration contribution 2014 AR5 SROCC Published range after 2100 Thermal 0.07-0.46 0.19--0.35 expansion 0.28-- Greenland 0.14 0.22--0.39 0.11--0.25 1.28 0.71-- -0.11-- Antarctica 0.21-0.25 -0.07--1.13 -0.14--0.78 1.35 1.56 Glaciers – 0.13--0.30 Land water 0.07- -0.03 0.05--0.10 storage 0.37 0.38- 0.57- Total (2300) 0.3--2.9 0.8--2.0 0.6--1.5 1.4--2.1 1.0-3.1 0.82 1.04 High RCP 8.5 SSP5-8.5 Post-AR5 Assessed Published range No ice-sheet ice-sheet MICI SEJ m rel. to 1995- without (with) acceleration contribution 2014 AR5 SROCC MICI after 2100 Thermal 0.28-1.80 expansion 0.91--1.50 0.40-- Greenland 0.30-1.18 0.53--0.89 0.32--1.74 2.23 0.02- 0.60- 6.87-- 0.03-- Antarctica 0.19 2.89 -0.24--1.68 -0.27--3.17 13.54 3.05 Do Not Cite, Quote or Distribute 9-120 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI Glaciers 0.29-0.39 0.32 Land water – storage 0.05--0.10 0.89- 2.25- 1.7--6.8 (up to Total (2300) 3.56 5.34 14.1) 1.9--4.1 2.2--5.9 9.5--16.2 2.4--6.3 1 2 [END TABLE 9.11 HERE] 3 4 5 On still longer timescales, the AR5 concluded with low confidence that the multi-millennial GMSL 6 commitment sensitivity to warming was about 1 to 3 m°C-1 GSAT increase. Two process-model-based 7 studies since the AR5 (Clark et al., 2016; Van Breedam et al., 2020) indicate higher commitments (Figure 8 9.30). Ice sheets dominate the multi-millennial sea-level commitment (Sections 9.4.1.4, 9.4.2.6), but the two 9 studies disagree on the relative contribution of the Greenland and Antarctic ice sheets. Notably, processes 10 such as Marine Ice Cliff Instability (Section 9.4.2.4) that are a major factor behind the deep uncertainty in 11 century-scale Antarctic ice sheet response do not appear to have a substantial effect on the multimillennial 12 magnitude (DeConto and Pollard, 2016). Only one of the studies of multimillennial GMSL commitments 13 includes scenarios consistent with 1.5°C of peak warming (Clark et al., 2016); this study suggests a 2000- 14 year commitment at 1.5°C of about 2.3-3.1 m, with approximately an additional 1.4-2.3 m commitment 15 between 1.5°C and 2.0°C (i.e., about 3 to 5 m °C-1). Taken together, both studies show a 2000-year GMSL 16 commitment of about 2-6 m for peak warming of about 2°C, 4-10 m for 3°C, 12-16 m for 4°C, and 19-22 m 17 for 5°C (medium agreement, limited evidence, Table 9.10). GMSL rise continues after 2000 years, leading to 18 a 10,000-year commitment of about 6-7 m for 1.5°C of peak warming (based on (Clark et al., 2016)), and of 19 about 8-13 m for 2.0°C, 10-24 m for 3.0°C, 19-33 m for 4.0°C, and 28-37 m for 5°C (based on both studies) 20 (medium agreement, limited evidence, Table 9.10). 21 22 23 [START FIGURE 9.30 HERE] 24 25 Figure 9.30: Global mean sea-level commitment as a function of peak global surface air temperature. From 26 models (Clark et al., 2016; DeConto and Pollard, 2016; Garbe et al., 2020; Van Breedam et al., 2020) and 27 paleo data on 2000-year (lower row) and 10,000 year (upper row) timescales. Columns indicate different 28 contributors to GMSL rise (from left to right: total GMSL change, Antarctic Ice Sheet, Greenland Ice 29 Sheet, global mean thermosteric sea-level rise, and glaciers). Further details on data sources and 30 processing are available in the chapter data table (Table 9.SM.9). 31 32 [END FIGURE 9.30 HERE] 33 34 35 An indicative metric for the equilibrium sea-level response can be provided by comparing paleo global 36 surface air temperature and GMSL during past multimillennial warm periods (Sections 2.3.1.1, 2.3.3.3 9.6.2, 37 Figure 9.9). However, caution is needed as the present and past warm periods differ in astronomical and 38 other forcings (Cross-chapter Box 2.1) and in terms of polar amplification. The Last Interglacial (likely 5-10 39 m higher GMSL than today and 0.5-1.5°C warmer than 1850-1900; Section 9.6.2; Table 9.6) is consistent 40 with the (Clark et al., 2016) projections for the 10,000-year commitment associated with 1.5°C of warming. 41 Similarly, the Mid-Pliocene Warm Period (very likely 5-25 m higher GMSL than today and very likely 2.5- 42 4°C warmer; Section 9.6.2; Table 9.6) is consistent with the range of 10,000 year commitments associated 43 with 2.5-4°C of warming, but GMSL reconstructions provide only a weak, broad constraint on model-based 44 projections. An additional paleo constraint comes from the Early Eocene Climatic Optimum, which indicates 45 that 10-18°C of warming is associated with ice-free conditions and a likely GMSL rise of 70-76 m (Section 46 2.3.3, Section 9.6.2). Together with model-based projections (Clark et al., 2016; Van Breedam et al., 2020), 47 this period suggests that commitment to ice-free conditions would occur for peak warming of about 7 - 13°C 48 (medium agreement, limited evidence). 49 Do Not Cite, Quote or Distribute 9-121 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 On the basis of modeling studies, paleo constraints, single-ice sheet studies finding multimillennial nonlinear 2 responses from both the Greenland and Antarctic ice sheets (Sections 9.4.1.4, 9.4.2.6), and the underlying 3 physics, we conclude that GMSL commitment is nonlinear in peak warming on timescales of both 2,000 and 4 10,000 years (medium confidence) and exceeds the AR5 assessment of 1 to 3 m °C-1 (medium agreement, 5 limited evidence) (Table 9.9). Although thermosteric sea level will start to decline slowly about 2,000 years 6 after emissions cease, the slower responses from the Greenland and Antarctic ice sheets mean that GMSL 7 will continue to rise for 10,000 years under most scenarios (medium confidence). 8 9 Since the AR5, a small number of modelling studies have examined the reversibility of the multimillennial 10 sea-level commitment under carbon dioxide removal, solar radiation modification or local ice-shelf 11 engineering. The slow response of the deep ocean to forcing leads to global-mean thermosteric sea-level fall 12 occurring long afterward even if CO2 levels are restored after a transient increase: global mean thermosteric 13 sea level takes over a millennium to reverse course (Ehlert and Zickfeld, 2018b). Rapid reversion to pre- 14 industrial CO2 concentrations has been found to be ineffective at fostering regrowth of the Antarctic ice sheet 15 (DeConto et al., 2021) but may reduce the multimillennial sea-level commitment (DeConto and Pollard, 16 2016). Altering sub-ice shelf bathymetry (Wolovick and Moore, 2018) or triggering ice-shelf advance 17 through massive snow deposition (Feldmann et al., 2019) might interrupt Marine Ice Sheet Instability 18 (Section 9.4.2.4) and thus reduce sea-level commitment. A reversion to pre-industrial Greenland ice sheet 19 temperatures with solar radiation modification is projected to stop mass loss in Greenland but leads to 20 minimal regrowth (Applegate and Keller, 2015). Based on limited evidence, carbon dioxide removal, solar 21 radiation modification, and local ice-shelf engineering may be effective at reducing the yet-to-be-realized 22 sea-level commitment but ineffective at reversing GMSL rise (low confidence). 23 24 25 [START BOX 9.4 HERE] 26 27 BOX 9.4: High-end storyline of 21st century sea-level rise 28 29 In this box, we outline a storyline (Glossary, Box 10.2 (Shepherd et al., 2018)) for high-end sea-level 30 projections for 2100. This storyline considers processes whose quantification is highly uncertain regarding 31 the timing of their possible onset and/or their potential to accelerate sea-level rise. These processes are 32 therefore not considered for the assessed upper bound of likely sea-level rise by 2100 in section 9.6.3.3, as 33 the likely range includes only processes that can be projected skilfully with at least medium confidence 34 (based on agreement and evidence). 35 36 As noted by the SROCC, stakeholders with a low risk tolerance (e.g., those planning for coastal safety in 37 cities and long-term investment in critical infrastructure) may wish to consider global-mean sea-level rise 38 above the assessed likely range by the year 2100, because “likely” implies an assessed likelihood of up to 16 39 % that sea-level rise by 2100 will be higher (see also (Siegert et al., 2020)). Because of our limited 40 understanding of the rate at which some of the governing processes contribute to long-term sea-level rise, we 41 cannot currently robustly quantify the likelihood with which they can cause higher sea-level rise before 2100 42 (Stammer et al., 2019). 43 44 In light of such deep uncertainty, we employ a storyline approach in examining the potential for, and early- 45 warning signals of a high-end sea-level scenario unfolding within this century. In doing so, we note upfront 46 that the main uncertainty related to high-end sea-level rise is “when" rather than “if” it arises: the upper limit 47 of 1.02 m of likely sea-level range by 2100 for the SSP 5-8.5 scenario will be exceeded in any future 48 warming scenario on time scales of centuries to millennia (high confidence), but it is uncertain how quickly 49 the long-term committed sea level will be reached (Section 9.6.3.5). Hence, global-mean sea level might rise 50 well above the likely range before 2100, which is reflected by assessments of ice-sheet contributions based 51 on structured expert judgment (Bamber et al., 2019) leading to a 95th percentile of projected future sea-level 52 rise as high as 2.3 m in 2100 (Section 9.6.3.3). 53 54 A plausible storyline for such high-end sea-level rise in 2100 assumes a strong warming scenario (Section 55 4.8). The storyline considers faster-than-projected disintegration of marine ice shelves and the abrupt, Do Not Cite, Quote or Distribute 9-122 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 widespread onset of Marine Ice Cliff Instability (MICI) and Marine Ice Sheet Instability (MISI) in Antarctica 2 (Section 9.4.2.4), and faster-than-projected changes in both the surface mass balance and dynamical ice loss 3 in Greenland. While conceptual studies provide medium evidence of these processes, substantial 4 uncertainties and low agreement in quantifying their future evolution arise from limited process 5 understanding, limited availability of evaluation data, missing or crude representation in model simulations, 6 their high sensitivity to uncertain boundary conditions and parameters, and/or uncertain atmosphere and 7 ocean forcing (Sections 9.4.1.2; 9.4.2.2). 8 9 In Antarctica, high warming might lead to floating ice shelves starting to break up earlier than expected due 10 to processes not yet accounted for in ice-sheet models or in current climate models used to force ice-sheet 11 projections. Such processes include hydrofracturing driven by surface meltwater and increase in ocean 12 thermal forcing driven by ocean circulation changes (Sections 9.2.2.3, 9.2.3.2, 9.4.2.3) (Hellmer et al., 2012, 13 2017; Silvano et al., 2018; Hazel and Stewart, 2020). In particular, the Thwaites and Pine Island Glacier ice 14 shelves could potentially disintegrate this century, which might trigger MICI before 2100 (DeConto and 15 Pollard, 2016; DeConto et al., 2021). MISI could potentially develop earlier and faster than simulated by the 16 majority of models if fast flowing ice streams follow plastic, instead of currently assumed more viscous, 17 sliding laws (Sun et al., 2020). Oceanic feedbacks could drive high-end sea-level rise by changes in the 18 meltwater-driven overturning circulation in ice cavities that cause additional melting (Jeong et al., 2020); by 19 a warming of the ocean water in contact with the ice shelves due to increased stratification and thus reduced 20 vertical mixing (Sections 9.2.2.3, 9.2.3.2) (Golledge et al., 2019b; Moorman et al., 2020; Sadai et al., 2020); 21 or by an increase in sea-ice cover due to increased ocean stratification (Section 9.3.2.1), which could reduce 22 the amount of warm, moist air that reaches the continent and limit the mass gain from snowfall over the ice 23 sheet (Sadai et al., 2020). 24 25 In Greenland, stronger mass loss than currently projected might also occur (Aschwanden et al., 2019; Khan 26 et al., 2020; Slater et al., 2020b). For example, warming-induced dynamical changes in atmospheric 27 circulation could enhance summer blocking and produce more frequent extreme melt events over Greenland 28 similar to the record mass loss of more than 500 Gt in summer 2019 (Section 9.4.1.1) (Delhasse et al., 2018; 29 Sasgen et al., 2020). Cloud processes in polar areas that are not well represented in models could further 30 enhance surface melt (Hofer et al., 2019), as could feedbacks between surface melt and the increasing albedo 31 from meltwater, detritus and pigmented algae (Section 9.4.1.1) (Cook et al., 2020). The same ice dynamical 32 processes associated with basal melt and MISI discussed for Antarctica could also occur in Greenland as 33 long as the ice sheet is in contact with the ocean. 34 35 The strength of all these processes is currently understood to depend strongly on global mean temperature 36 and polar amplification, with additional linkages through feedback from global mean sea-level (Gomez et al., 37 2020). These dependencies on a joint forcing imply that processes are strongly correlated. Hence, both their 38 uncertainties and their possible cascading contribution to high-end sea-level rise are expected to combine. 39 High-end sea-level rise can therefore occur if one or two processes related to ice-sheet collapse in Antarctica 40 result in an additional sea-level rise at the maximum of their plausible ranges (Sections 9.4.2.5, 9.6.3.3; 41 Table 9.7) or if several of the processes described in this box result in individual contributions to additional 42 sea-level rise at moderate levels. In both cases, global-mean sea-level rise by 2100 would be substantially 43 higher than the assessed likely range, as indicated by the projections including low confidence processes 44 reaching in 2100 as high as 1.6 m at the 83rd percentile and 2.3 m at the 95th percentile (Section 9.6.3.3). 45 46 Identifying the potential drivers of a high-end sea-level rise allows identification of sites and observables that 47 can provide early warnings of a much faster sea-level rise than the likely range of this and previous reports. 48 One potential site for such monitoring is Thwaites Glacier, which is melting faster in some places and slower 49 in others than models simulate. At this glacier, the effect of tides and channelling of warm water flows on the 50 melting is evident (Milillo et al., 2019), making the floating ice shelf potentially vulnerable to breakup from 51 hydrofracturing, driven by surface meltwater, much earlier than expected. In addition, the glacier is 52 retreating towards a zone with deeper bedrock, which at its present rate of retreat would be reached in 30 53 years (Yu et al., 2019). Thwaites Glacier is therefore a strong candidate to experience large-scale MISI 54 and/or MICI (Golledge et al., 2019b; DeConto et al., 2021), making it the ideal site for monitoring early- 55 warning signals of accelerated sea-level-rise from Antarctica. Such signals could possibly be observed within Do Not Cite, Quote or Distribute 9-123 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 the next few decades (Scambos et al., 2017). 2 3 [END BOX 9.4 HERE] 4 5 6 9.6.4 Extreme sea levels: Tides, surges and waves 7 8 An extreme sea level (ESL) refers to an occurrence of exceptionally high or low local sea surface height 9 (Box 9.1). This section focusses on oceanographic driven changes in ESL (Box 9.1). 10 11 12 9.6.4.1 Past changes 13 14 The AR5 (Church et al., 2013a) concluded that changes in extreme still water levels (ESWL, combining 15 relative sea level, tide and surge as observed by tide gauges: Box 9.1) are very likely to be caused by 16 observed increases in relative sea level, but noted low confidence in region-specific results owing to the 17 limited number of studies considering localised contributions from storm surge, tide or wave effects. 18 Influences from dominant modes of climate variability, particularly ENSO and NAO (Annex IV), were also 19 noted. Climate modes affect sea level extremes in many regions, as a result of both sea-level anomalies 20 (Sections 9.2.4.2, 9.6.1.3) and changes in storminess (Section 11.7). The SROCC (Oppenheimer et al., 2019) 21 concluded with high confidence that inclusion of local processes (wave effects, storm surges, tides plus other 22 regional morphology changes due to erosion, sedimentation and compaction) is essential for estimation of 23 changes in ESL events. 24 25 As in the AR5 and the SROCC, tide gauge observations show that relative sea level rise (Section 9.6.1.3) is 26 the primary driver of changes in ESWL at most locations and, across tide gauges, has led to a median 165% 27 increase in high-tide flooding over 1995-2014 relative to those over 1960-1980 (Figure 9.31) (high 28 confidence). Some locations exhibit substantial differences between long-term relative sea level trends and 29 ESWL (high confidence), particularly given decadal to multidecadal variations of other ESWL contributors 30 (Rashid and Wahl, 2020). Since the SROCC, relative sea level rise has been shown to be the dominant 31 contributor to ESWL rise at most gauge sites along the Chinese coast, but, at some locations, the surge 32 contribution dominates (Feng et al., 2019). Trends in the difference between ESWL and mean relative sea 33 level rise can result from changes (either positive or negative) in the surge or tidal components, and can 34 include non-linear interactions between tide, surge, and relative sea level (Arns et al., 2015; Schindelegger et 35 al., 2018). The positive phase of the 18.6 year nodal cycle of the astronomical tide is a further consideration, 36 contributing to an increased flood hazard relative to the long term average (Talke et al., 2018; Peng et al., 37 2019; Baranes et al., 2020). Failing to consider the non-linear interactions between tide, surge and relative 38 sea level may overestimate trends in ESWL (Arns et al., 2020) (low confidence), and, in some regions, 39 changes in ESWL depend more on changes in surge or tide than on sea level trends. 40 41 Ongoing development of the GESLA tide gauge database (Woodworth et al., 2016) along with data 42 archaeology (Talke and Jay, 2013) extends availability of tide gauge records back to the mid 19th Century (or 43 earlier). Dynamical datasets used to assess trends in ESL at global or regional scales (e.g., tide and surge 44 contributions from the Global Tide and Surge Reanalysis (GTSR) (Muis et al., 2016, 2020), or wave 45 setup/swash contributions from available wave hindcasts/reanalyses (Melet et al., 2018)) have model biases 46 introduced with resolution and parameterisation limitations, incomplete atmospheric data and currently span 47 only a few decades, so they are not yet long or accurate enough to assess long-term trends in ESLs. 48 Therefore, there is medium confidence in observed trends in ESWL but only low confidence in modelled 49 ESL trends. 50 51 The AR5 indicated that the amplitude and phase of major tidal constituents have exhibited long-term change, 52 but that their effects on ESL were not well understood. The SROCC (Bindoff et al., 2019) reported changes 53 in tides (amplification and dampening) at some locations to be of comparable importance to changes in mean 54 sea level for explaining changes in high water levels, with the sign of change being dependent on stability of 55 shoreline position. Relative sea level rise causes water depth-based alterations to the resonant characteristics Do Not Cite, Quote or Distribute 9-124 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 of the basin, changes the bottom friction and increases the wave speed (Pickering et al., 2012) and remains 2 the primary hypothesis for observed tidal changes. Other contributing processes include strong localised 3 anthropogenic drivers (e.g., port development, dredging, flood defences, land reclamation), changes in 4 stratification associated with ocean warming (Section 9.2.1.3), and changes in seabed roughness associated 5 with ecological change (e.g., Haigh et al., 2019). Tide gauge data show that, although principal tidal 6 components have varied in amplitude on the order of 2% to 10% per century (Jay, 2009; Ray, 2009), 7 identifying direct causality remains challenging (Haigh et al., 2019). Combined, observations and models 8 indicate relative sea level rise and direct anthropogenic factors are the primary drivers of observed tidal 9 changes at tide gauge stations (medium confidence). 10 11 The SROCC (Oppenheimer et al., 2019) reported variations in storm surge not related to changes in relative 12 sea level, and concluded with high confidence that consideration of localised storm surge processes was 13 essential to monitor trends in ESL. Storm surge-driven ESL events are a response to tropical and 14 extratropical cyclones. While historical trends in extra-tropical cyclones are less clear (Section 11.7.2.1), 15 there is mounting evidence for an increasing proportion of stronger tropical cyclones globally, with an 16 associated poleward migration (Section 11.7.1.2). These changes are captured in the ESL record, for 17 example, via increasing intensity and poleward shift in the location of typhoon-driven storm surges reported 18 across 64 years (1950-2013) in the western North Pacific (Oey and Chou, 2016). Along the US east coast, 19 there has been an increase in frequency of ESL events due to tropical cyclone changes since 1923 that can be 20 statistically linked to changes in global average temperature (Grinsted et al., 2013), and the signal is 21 projected to emerge around 2030 (Lee et al., 2017). At century and longer timescales, geological proxies 22 such as overwash deposits in coastal lagoons or sinkholes can be used to reconstruct past changes in storm 23 activity (e.g., Brandon et al., 2013; Lin et al., 2014) and put recent events into historical perspective (e.g., 24 Brandon et al., 2015). However, there is low confidence in the current ability to quantitatively compare 25 geological proxies with gauge data. Historical storm surge activity is being increasingly assessed with use of 26 hydrodynamic model simulations and data-driven global reconstructions to supplement tide-gauge 27 observations to investigate historical changes at centennial to millennial time scales (e.g., (Ji et al., 2020; 28 Muis et al., 2020; Tadesse et al., 2020). Large regional variations and limited observational data lead to low 29 confidence in observed trends in the surge contribution to increasing ESL. 30 31 32 [START FIGURE 9.31 HERE] 33 34 Figure 9.31: Historical occurrences of minor extreme still water levels. Defined as the 99th percentile of daily 35 observed water levels over 1995-2014. (a) Percent change in occurrences over 1995-2014 relative to 36 those over 1960-1980. (b-g) Annual mean sea level (blue) and annual occurrences of extreme still water 37 levels over the 1995-2014 99th percentile daily maximum (yellow) at six selected tide gauge locations. 38 Further details on data sources and processing are available in the chapter data table (Table 9.SM.9). 39 40 [END FIGURE 9.31 HERE] 41 42 43 Waves contribute to ESL via wave setup, infra-gravity waves and swash processes (Dodet et al., 2019), with 44 Extreme Total Water Level (ETWL: Box 9.1) used to represent ESWL with addition of wave setup, and 45 Extreme Coastal Water Level (ECWL: Box 9.1) also including contributions from swash. The SROCC 46 (Oppenheimer et al., 2019) reported the dependency of these processes on nearshore geomorphology and 47 deep-water wave climate, and thus sensitivity to internal climate variability and climate change. Few long 48 term deployments of in-situ measurements in the very dynamic surf zone means that long term records of 49 ETWL or ECWL are limited to a few sites; tidal gauges are typically located in sheltered locations (e.g., 50 harbours) where wave contributions are absent (Lambert et al., 2020). Consequently, trends in wave 51 contributions to ESL are typically derived from trends in wave conditions observed offshore. On the basis of 52 satellite altimeter observations, the SROCC reported increasing extreme wave heights in the Southern and 53 North Atlantic Oceans of around 1.0 cm yr-1 and 0.8 cm yr-1 over the period 1985-2018 (medium confidence). 54 The SROCC (Collins et al., 2019) also identified sea-ice loss in the Arctic as leading to increased wave 55 heights over the period 1992 to 2014 (medium confidence). Since the SROCC, the satellite wave record has Do Not Cite, Quote or Distribute 9-125 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 been shown to be sensitive to alternate processing techniques, leading to important differences in reported 2 trends (Timmermans et al., 2020). The most common observation platform for surface waves over the past 3 30 years are in-situ buoys. However, evolving biases associated with changing instrument type, 4 configuration and sampling methodology introduce artificial trends (e.g., Gemmrich et al., 2011; 5 Timmermans et al., 2020). Accurate metadata is required to address these issues, and, while available 6 locally, are only beginning to be globally coordinated (Centurioni et al., 2019). Wave reanalysis and hindcast 7 products have also been used to investigate total water level at global scale (Melet et al., 2018; Reguero et 8 al., 2019). Their applicability for trend analysis is limited by inhomogeneous data for assimilation (Stopa et 9 al., 2019), but they inform relationships between seasonal, inter-annual to inter-decadal variability of climate 10 indices and wind-wave characteristics (Marshall et al., 2015a, 2018; Kumar et al., 2016; Stopa et al., 2016). 11 To summarise, satellite era trends in wave heights of order 0.5 cm yr-1 have been reported, most pronounced 12 in the Southern Ocean. However, sensitivity of processing techniques, inadequate spatial distribution of 13 observations, and homogeneity issues in available records limit confidence in reported trends (medium 14 confidence). 15 16 Only a few studies have attempted to quantify the role of anthropogenic climate change in ESL events (e.g. 17 Mori et al., 2014, Takayabu et al., 2015, Turki et al., 2019). Detection and attribution of the human influence 18 on climatic changes in surges, and waves remains a challenge (Ceres et al., 2017), with limited evidence to 19 suggest in some instances (e.g., poleward migration of tropical cyclones in the Western North Pacific: 20 Section 11.7.1.2), changes in surges and waves can be attributed to anthropogenic climate change (low 21 confidence). With relative sea-level change being considered the primary driver of observed tidal changes, 22 there is medium confidence that these changes can be attributed to human influence. The close relationship 23 between local ESL and long-term relative sea level change, combined with the robust attribution of GMSL 24 change (Section 9.6.1.4), implies that observed global changes in ESL can be attributed, at least in part, to 25 human-caused climate change (medium confidence), but reconciling regional variation in these changes is 26 not yet possible (Section 9.6.1.4). 27 28 29 9.6.4.2 Future changes 30 31 There are two distinct methods used to project future ESL changes. The static, or mean sea level offset, 32 approach employs historical distributions of tidal, surge and wave components and adjusts future ESL 33 distributions for mean relative sea level rise. The dynamic approach employs hydrodynamic and/or wave 34 models forced with GCM-derived atmospheric fields to project changes in tidal, storm surge and wave 35 distributions, which are then combined with relative sea level projections to project future ESLs. The 36 dynamic approach is computationally expensive. Use of the dynamic approach on large spatial or global 37 scales has only recently been successful to project 21st Century changes in ETWL (Vousdoukas et al., 2017, 38 2018) and ECWL (Melet et al., 2020). (Kirezci et al., 2020) assume stationarity in global wave and storm 39 surge simulations to assess projected 21st century changes in episodic coastal ETWL driven flooding under 40 global sea-level rise scenarios. 41 42 The SROCC (Oppenheimer et al., 2019) presents projections of ESL derived using a static approach. Such 43 projections often quantify changes in ESL event frequency, expressed as “frequency amplification factors” 44 (Hunter, 2010, 2012). Like relative sea level projections, frequency amplification factors increase under 45 higher emission scenarios, and differences between scenarios increase over time. The SROCC concludes that 46 even small to moderate changes in mean relative sea level can lead to hundred- to thousand-fold increases in 47 the frequencies with which certain thresholds are exceeded; e.g., what is currently a 1-in-100 year ESL 48 height (1% annual probability or 0.01 expected annual events) will be expected once or even multiple times 49 per year in future at many locations (Figure 9.32). The SROCC showed that currently rare ESL events (e.g., 50 with an average return period of 100 years) will occur annually or more frequently at most available 51 locations for RCP4.5 by the end of the century (high confidence). Results from these assessments are 52 sensitive to the type of ESL probability distribution assumed (Buchanan et al., 2016; Wahl et al., 2017), as 53 well as the magnitude and uncertainty of projected relative sea level change (Slangen et al., 2017; Wahl et 54 al., 2017; Frederikse et al., 2020a). Frequency amplification factors tend to be largest in tropical regions due 55 in part to higher relative sea level rise projections, but primarily to the relative rarity of high ESLs in areas Do Not Cite, Quote or Distribute 9-126 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 with little historical exposure to tropical or extratropical cyclones. Alternative representation of changes in 2 ESL, such as presenting changes in exceedances per year (Sweet and Park, 2014), are subject to similar 3 sensitivities, and lead to medium confidence in projected changes of event frequency using these methods. 4 5 Employing a similar static approach (fitting a Gumbel distribution between Mean Higher High Water 6 (average of higher high water height of each tidal day) and a threshold following (Buchanan et al., 2016)), 7 this report updates the SROCC projections of ESL with the relative sea level projections from Section 8 9.6.3.3 (see also Supplementary Material 9.SM.4). By 2050, the median increase in frequency amplification 9 factor at 634 tide gauge stations is 19 for SSP1-2.6, 22 for SSP2-4.5 and 30 for SSP5-8.5 (Figure 9.32). This 10 means that by 2050 a historical (1995-2014) 1% annual probability ESL will have increased to an 19-30% 11 annual probability. The 1% historical annual probability event is expected to become an annual event at 19- 12 31% of the 634 stations by 2050, consistent with the SROCC. By 2100, the median frequency amplification 13 factor is projected to be 163 for SSP1-2.6, 325 for SSP2-4.5 and 532 for SSP5-8.5, with respectively 60%, 14 71%, and 82% of the stations experiencing a currently 1% annual probability event at least yearly (Figure 15 9.32) (medium confidence). 16 17 In the dynamic approach, the low resolution of the forcing fields arising from GCMs limits the ability to 18 resolve historical and future changes in tropical and extra-tropical storm frequency and intensity, and 19 resolution of local geography and morphology limit ability to represent ECWL (Box 9.1). Not all relevant 20 processes, e.g., river discharge, are included in the dynamic models, and ESL events are typically a 21 combination of multiple contributing processes, which are often not independent (Jevrejeva et al., 2019). In 22 both static and dynamical approaches, global assessment of the performance of modelled storm surge and 23 wave contributions to ESL is limited by poor coverage of observations (limited to tide-gauges for ESWL, 24 (Muis et al., 2020), and unavailable for the wave dependent ETWL and ECWL estimates (Vitousek et al., 25 2017; Vousdoukas et al., 2018; Kirezci et al., 2020; Lambert et al., 2020; Melet et al., 2020). In studies to 26 date, individual models are used to simulate different contributions to ESL, non-linear interactions are not 27 well captured, and uncertainties associated with downscaling methodology are poorly resolved, leading to 28 low confidence in available ESL projections that include these modelled wave and surge contributions. 29 30 Assessment of dynamic ETWL changes for regions is presented in Chapter 12, following the methods of 31 (Vousdoukas et al., 2018) and (Kirezci et al., 2020). Consistent with studies using the static approach, 32 (Vousdoukas et al., 2018) finds that by 2050 the historical 1% average annual probability ETWL will have 33 increased to a 2-50% average annual probability for most high latitude regions, and more often (up to 34 multiple times a year, >100% annual probability) in the tropics, under both RCP 4.5 and RCP 8.5. For 2100, 35 present-day 1% average annual probability extreme sea levels will be exceeded multiple times each year 36 almost everywhere. In summary, despite waves and surges being non-negligible contributors to projected 37 ETWL and ECWL changes (Vousdoukas et al., 2018; Melet et al., 2020), relative sea level change is 38 expected to be the main driver in changes in future ESL return periods in most areas (medium confidence). 39 40 The SROCC (Bindoff et al., 2019) concluded that the majority of coastal regions will experience statistically 41 significant changes in tidal amplitudes through the 21st Century. Comprehensive high-resolution (of order 42 10km) numerical modelling studies provide evidence for spatially coherent changes in tidal amplitudes in 43 shelf seas as a result of relative sea level rise (Haigh et al., 2019, and references therein). There is high 44 confidence that GMSL rise will be the primary driver of global tidal amplitude increases and decreases over 45 the next 100-200 years, changing the baseline tide that ESLs are imposed upon. At local and regional scales, 46 anthropogenic factors such as major land reclamation efforts (e.g., East China Sea, Song et al., 2013) or 47 differing national coastal management strategies (maintaining the present coastline position or managed 48 retreat) will locally modulate the influence of GMSL rise on tidal amplitude (medium confidence). 49 50 The SROCC (Oppenheimer et al., 2019) concluded that the intensity of severe tropical cyclones will increase 51 in a warmer climate (Section 11.7.1), but low confidence remains in the future frequency of tropical 52 cyclones. Changes in tropical cyclone climatology will contribute to variations in frequency and magnitude 53 of future ESL surge events, although estimates of this contribution range widely (Lin et al., 2012; McInnes et 54 al., 2014, 2016; Little et al., 2015; Garner et al., 2017; Mori et al., 2019; Muis et al., 2020). In the Gulf of 55 Mexico, changes in ESL due to tropical cyclone activity may be as important as SLR in enhancing future Do Not Cite, Quote or Distribute 9-127 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 flood hazards (Marsooli et al., 2019). For the Korean Peninsula, a maximum change in 100-year return 2 height associated with typhoon-induced storm surges of 10% under 4°C warming is found (Yang et al., 3 2018). The effects of projected changes in tropical cyclone intensity may be enhanced or offset in different 4 locations by effects of changes in tracks (Garner et al., 2017) (Section 11.7.1). There is low confidence in 5 projected changes in ESL driven by changes in tropical cyclone climatology. 6 7 Changes in surface wave conditions occur in response to changes in frequency; intensity and position of 8 forcing winds and storms (Morim et al., 2018, 2019); reduction in sea-ice and associated changes in fetch 9 conditions (Thomson and Rogers, 2014; Casas‐Prat and Wang, 2020); and changes in coastal morphology 10 associated with relative sea level rise (Wandres et al., 2017; Storlazzi et al., 2018). A few studies considering 11 the contribution of a non-stationary wave climate on future changes in ESL infer a small but non-negligible 12 contribution (Vousdoukas et al., 2018; Melet et al., 2020). The SROCC presented qualitative assessments of 13 projected changes in wave conditions. Since the SROCC, a quantitative assessment of a community 14 ensemble of global wind-wave projections (Morim et al., 2019) found robust projected changes of ~5-10% 15 (positive or negative, depending on region) in annual mean significant wave height, mean wave period, 16 and/or mean wave directions along ~52% of the world’s coastline that exceed internal climate variability 17 under RCP8.5 by 2100. Continued retreat of sea-ice cover in the Arctic will lead to more energetic wind- 18 wave conditions (Casas‐Prat and Wang, 2020). Wave climate modelling methods introduce up to ~50% of 19 the ensemble variance in mean wave climate projections (Morim et al., 2019). GCMs do not typically 20 resolve the higher-resolution tropical and extratropical storm features required to accurately determine the 21 contribution of extreme waves to ESLs and individual studies have sought to improve resolution to address 22 these issues (e.g., Timmermans et al., 2017). To date, projections of wave height extremes have been 23 constrained to single wave model configurations (e.g., Timmermans et al., 2017; Meucci et al., 2020). In 24 summary, there is medium confidence in projections of changes in mean wave climate but low confidence in 25 the projected changes in extreme wave conditions due to limited evidence. 26 27 Correlations between changes in sea level-forced (mean sea level and tidal) and atmospherically-forced 28 drivers (ocean surface waves and surges) of ESLs have only been considered in a few studies, although high 29 surge and high waves co-occur along a majority of the world’s coastlines (Marcos et al., 2019). Along the 30 US east coast, ocean dynamic sea level change and change in power dissipation index (a proxy for North 31 Atlantic tropical cyclone activity) are correlated across CMIP5 GCMs, resulting in an increase in ESLs 32 relative to analyses assuming independence of these changes (Little et al., 2015). In the Irish Sea, 33 dynamically coupled wave-tide modelling results in high water wave heights up to 20% higher than in an 34 uncoupled analysis (Lewis et al., 2019). In the German Bight, relative sea level rise relaxes the breaking 35 criterion of nearshore waves (assuming no geomorphological response), allowing larger waves to propagate 36 closer to shore, leading to increased wave runup (Arns et al., 2017). In south-western Australia, the influence 37 of projected SLR was found to exceed the influence of projected changes in forcing winds on wave 38 characteristics at the coast (Wandres et al., 2017). Thus, projections of ESL that do not consider correlations 39 between and among sea level-forced and atmospherically-forced drivers can differ strongly from coupled 40 projections (medium confidence). 41 42 The SROCC (Collins et al., 2019b) highlighted compound events, or coincident occurrence of multiple 43 hazards, as an example of deep uncertainty, and noted that failing to account for multiple factors contributing 44 to extreme events will lead to underestimation of the probabilities of occurrence (high confidence). Statistical 45 studies have shown that high rain or streamflow often co-occurs with storm surge as examples of 46 “compound” surge-rain or surge-discharge events (Sections 11.8.1; 12.4.5.6; Wahl and Chambers, 2015; 47 Moftakhari et al., 2017; Ward et al., 2018; Wu et al., 2018; Couasnon et al., 2019). Dynamical modelling 48 studies show co-occurrence of flood drivers raises ESLs at some locations in estuaries (e.g., Rhine Delta, 49 Zhong et al., 2013; the Netherlands, van den Hurk et al., 2015; Taiwan, Chen and Liu, 2016; and the Hudson 50 River, USA Orton et al., 2018), particularly when hydrologic catchments are steep and cause high rainfall 51 near the coast (SW UK, Svensson and Jones, 2004). The compound effect of storm surge and rainfall 52 contributes greater projected flood risk than climate induced amplification (Hsiao et al., 2021). However, at 53 other locations, co-occurrence was unimportant because streamflow timing was not coincident with the 54 coastal peak storm surge (Hudson River, Orton et al., 2012; Rhine delta, Klerk et al., 2015). The SROCC 55 (Oppenheimer et al., 2019) detailed the complexity of interactions in deltaic environments. Direct increases Do Not Cite, Quote or Distribute 9-128 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 in flooding driven by increasing relative sea levels and by increased storm surge, rain, or correlations 2 between these flood-drivers (e.g., Moftakhari et al., 2017; Orton et al., 2018) are expected to be further 3 accompanied by increases in flooding due to subsidence (vertical land movement) and sedimentation 4 (relative sea level driven blockage of river flows). The probability of concurrent surge, wave and 5 precipitation events has been projected to increase by more than 25% by 2100 compared to present, with 6 high northern latitudes displaying compound flooding becoming more than 2.5 times as frequent, and 7 weakening in the subtropics (Bevacqua et al., 2020). However, the number of studies on compound events is 8 still limited and therefore there is low confidence in understanding the extent by which compound surge-rain 9 events will change in response to relative sea level rise and climate change. 10 11 12 [START FIGURE 9.32 HERE] 13 14 Figure 9.32: Projected median frequency amplification factors for the 1% average annual probability extreme 15 still water level in 2050 (a, c, e) and 2100 (b, d, f). Based on a peak-over-threshold (99.7%) method 16 applied to the historical extreme still water levels of GESLA2 following SROCC and additionally fitting 17 a Gumbel distribution between MHHW and the threshold following (Buchanan et al., 2016), using the 18 regional sea-level projections of this chapter (Section 9.6.3.3) for (a, b) SSP5-8.5, (c, d) SSP2-4.5 and (e, 19 f) SSP1-2.6. Further details on data sources and processing are available in the chapter data table (Table 20 9.SM.9). 21 22 [END FIGURE 9.32 HERE] 23 24 25 9.7 Final Remarks 26 27 The process-based assessment of observed and projected change in the ocean, cryosphere and sea level 28 undertaken here reveals advances and gaps in reconstructions, observations, models and process 29 understanding. Revisiting the updated assessments since the AR5 and the SROCC helps to gauge the 30 robustness of understanding and quantitative assessments. The CMIP6 family of models builds upon the 31 experience of the CMIP5 models, and the projections of ISMIP6, LARMIP2 and GlacierMIP strengthen 32 understanding. Taken together with emulators of these simulations (Box 9.3) and transparent statistical 33 approaches (Section 9.6.3), this chapter provides projections which are consistent with the assessment of 34 Equilibrium Climate Sensitivity in this report and that have improved estimates of uncertainty. 35 36 The largest uncertainties in future sea level and cryosphere change are related to the Greenland and Antarctic 37 ice sheets (Sections 9.4.1.3, 9.4.1.4, 9.4.2.5, 9.4.2.6). While the ISMIP6 and LARMIP2 protocols provide 38 simulations permitting uncertainty estimation and probabilistic inferences, remaining deep uncertainty 39 relates both to ice sheet processes and the atmospheric and oceanic conditions simulated by CMIP models in 40 polar regions (Sections 9.4.2.3, 9.4.2.4). ISMIP6 and LARMIP2 have not been simulated beyond 2100, 41 which greatly reduces the amount and variety of state-of-the-art projections available to make ice sheet and 42 sea level projections beyond 2150. After 2150, limited agreement causes us to consider all projections as low 43 confidence. Critically, the uncertainty in ice sheet projections is the leading uncertainty in projections of 44 future global sea level for the second half of this century and beyond (Section 9.6.3). 45 46 Glacier inventory and projection uncertainty has been a significant source of past sea level budget 47 uncertainty and remains a dominant uncertainty until mid-century. Emission scenario becomes the largest 48 source of glacier change uncertainty by 2100 just as the relative importance of glacier loss is projected to 49 decrease (Section 9.5.1). 50 51 New high-resolution climate models show that SST, overturning circulation, ocean heat content change and 52 sea-ice cover are considerably improved in most models when compared to the coarser resolution models. 53 Change in the Southern Ocean and adjacent shelves (Section 9.2.3.2) is intimately linked to the future of the 54 Antarctic ice sheet (Section 9.4.2.3), and projection of the Southern Ocean depends on both oceanic and 55 atmospheric drivers affecting heat (and carbon) uptake and sea ice. However, resolution remains a factor, as 56 most CMIP6 models are far from resolutions that directly represent coastal and regional shallow-water Do Not Cite, Quote or Distribute 9-129 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 processes such as those beneath Antarctic ice shelves, in Greenland fjords and the eddying convection found 2 by OSNAP. 3 4 Processes that change on long timescales—particularly AMOC, ocean heat content, and ice sheets—require 5 additional projections beyond the CMIP scenarios to explore longer term commitment, post-forcing recovery 6 measured in centuries rather than years or decades, and potential tipping points and thresholds. There were 7 only a few new studies focussed on longer timescales and none based on CMIP6 models. 8 9 Do Not Cite, Quote or Distribute 9-130 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 Frequently Asked Questions 2 3 FAQ 9.1: Can continued melting of the Greenland and Antarctic ice sheets be reversed? How long 4 would it take for them to grow back? 5 6 Evidence from the distant past shows that some parts of the Earth system might take hundreds to thousands 7 of years to fully adjust to changes in climate. This means that some of the consequences of human-induced 8 climate change will continue for a very long time, even if atmospheric heat-trapping gas levels and global 9 temperatures are stabilized or reduced in the future. This is especially true for the Greenland and Antarctic 10 ice sheets, which grow much more slowly than they retreat. If the current melting of these ice sheets 11 continues for long enough it becomes effectively irreversible on human timescales, as does the sea level rise 12 caused by that melting. 13 14 Humans are changing the climate and there are mechanisms that amplify the warming in the polar regions 15 (Arctic and Antarctic). The Arctic is already warming faster than anywhere else (see FAQ 4.3). This is 16 significant because these colder high latitudes are home to our two remaining ice sheets: in Antarctica and 17 Greenland. Ice sheets are huge reservoirs of frozen freshwater, built up by tens of thousands of years of 18 snowfall. If they were to completely melt, the water released would raise global sea level by about 65 m. 19 Understanding how these ice sheets are affected by warming of nearby ocean and atmosphere is therefore 20 critically important. The Greenland and Antarctic ice sheets are already slowly responding to recent changes 21 in climate, but it takes a long time for these huge masses of ice to adjust to changes in global temperature. 22 That means that the full effects of a warming climate may take hundreds or thousands of years to play out. 23 An important question is whether these changes can eventually be reversed, once levels of greenhouse gases 24 in the atmosphere are stabilized or reduced by humans and natural processes. Records from the past can help 25 us answer this question. 26 27 For at least the last 800,000 years, the Earth has followed cycles of gradual cooling followed by rapid 28 warming caused by natural processes. During cooling phases, more and more ocean water is gradually 29 deposited as snowfall, causing ice sheets to grow and sea level to slowly decrease. During warming phases, 30 the ice sheets melt more quickly, resulting in more rapid rises in sea level (FAQ 9.1, Figure 1). Ice sheets 31 build up very slowly because growth relies on the steady accumulation of falling snow that eventually 32 compacts into ice. As the climate cools, areas that can accumulate snow expand, reflecting back more 33 sunlight that otherwise would keep the Earth warmer. This means that once started, glacial climates develop 34 rapidly. However, as the climate cools, the amount of moisture that the air can hold tends to decrease. As a 35 result, even though glaciations begin quite quickly, it takes tens of thousands of years for ice sheets to grow 36 to a point where they are in balance with the colder climate. 37 38 Ice sheets retreat more quickly than they grow because of processes that, once triggered, drive self- 39 reinforcing ice loss. For ice sheets that are mostly resting on bedrock above sea level – like the Greenland ice 40 sheet – the main self-reinforcing loop that affects them is the ‘elevation–mass balance feedback’ (FAQ 9.1, 41 Figure 1, right). In this situation, the altitude of the ice sheet surface decreases as it melts, exposing the sheet 42 to warmer air. The lowered surface then melts even more, lowering it faster still, until eventually the whole 43 ice sheet disappears. In places where the ice sheet rests instead on bedrock that is below sea level and which 44 also deepens inland, including many parts of the Antarctic ice sheet, an important process called ‘marine ice- 45 sheet instability’ is thought to drive rapid retreat (FAQ 9.1, Figure 1, left). This happens when the part of the 46 ice sheet that is surrounded by sea water melts. That leads to additional thinning, which in turn accelerates 47 the motion of the glaciers that feed into these areas. As the ice sheet flows more quickly into the ocean, more 48 melting takes place, leading to more thinning and even faster flow that brings ever-more glacier ice into the 49 ocean, ultimately driving rapid deglaciation of whole ice-sheet drainage basins. 50 51 These (and other) self-reinforcing processes explain why relatively small increases in temperature in the past 52 led to very substantial sea level rise over centuries to millennia, compared to the many tens of thousands of 53 years it takes to grow the ice sheets that lowered the sea level in the first place. These insights from the past 54 imply that if human-induced changes to the Greenland and Antarctic ice sheets continue for the rest of this 55 century, it will take thousands of years to reverse that melting, even if global air temperatures decrease Do Not Cite, Quote or Distribute 9-131 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 within this or the next century. In this sense, these changes are therefore irreversible, since the ice sheets 2 would take much longer to regrow than the decades or centuries for which modern society is able to plan. 3 4 5 [START FAQ9.1, FIGURE 1 HERE] 6 7 FAQ 9.1, Figure 1: Ice sheets growth and decay. (Top) Changes in ice-sheet volume modulate sea level variations. 8 The grey line depicts data from a range of physical environmental sea-level recorders such as coral 9 reefs (see Table 9.SM.5) while the blue line is a smoothed version of it. (Bottom, left) Example of 10 destabilisation mechanism in Antarctica. (Bottom, right) Example of destabilisation mechanism in 11 Greenland. 12 13 14 [END FAQ9.1, FIGURE 1 HERE] 15 16 17 Do Not Cite, Quote or Distribute 9-132 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 FAQ 9.2: How much will sea level rise in the next few decades? 2 3 As of 2018, global average sea level was about 15–25 cm higher than in 1900 and 7–15 cm higher than in 4 1971. Sea level will continue to rise by an additional 10–25 cm by 2050. The major reasons for this ongoing 5 rise in sea level are the thermal expansion of seawater as its temperature increases and the melting of 6 glaciers and ice sheets. Local sea level changes can be larger or smaller than the global average, with the 7 smallest changes in formerly glaciated areas and the largest changes in low-lying river delta regions. 8 9 Across the globe, sea level is rising, and the rate of increase has accelerated. Sea level increased by about 4 10 mm per year from 2006 to 2018, which was more than double the average rate over the 20th century. Rise 11 during the early 1900s was due to natural factors, such as glaciers catching up to warming that occurred in 12 the Northern Hemisphere during the 1800s. However, since at least 1970, human activities have been the 13 dominant cause of global average sea level rise, and they will continue to be for centuries into the future. 14 15 Sea level rises either through warming of ocean waters or the addition of water from melting ice and bodies 16 of water on land. Expansion due to warming caused about 50% of the rise observed from 1971 to 2018. 17 Melting glaciers contributed about 22% over the same period. Melting of the two large ice sheets in 18 Greenland and Antarctica has contributed about 13% and 7%, respectively, during 1971-2018, but melting 19 has accelerated in the recent decades, increasing their contribution to 22% and 14% since 2016. Another 20 source is changes in land water storage: reservoirs and aquifers on land have reduced, which contributed 21 about a 8% increase in sea level. 22 23 By 2050, sea level is expected to rise an additional 10–25 cm whether or not greenhouse gas emissions are 24 reduced (FAQ 9.2, Figure 1). Beyond 2050, the amount by which sea level will rise is more uncertain. The 25 accumulated total emissions of greenhouse gases over the upcoming decades will play a big role beyond 26 2050, especially in determining where sea level rise and ice sheet changes eventually level off. 27 28 Even if net zero emissions are reached, sea level rise will continue because the deep ocean will continue to 29 warm and ice sheets will take time to catch up to the warming caused by past and present emissions: ocean 30 and ice sheets are slow to respond to environmental changes (see FAQ 5.3). Some projections under low 31 emissions show sea level rise continuing as net zero is approached at a rate comparable to today (3–8 mm 32 per year by 2100 versus 3–4 mm per year in 2015), while others show substantial acceleration to more than 33 five times the present rate by 2100, especially if emissions continue to be high and processes that accelerate 34 retreat of the Antarctic Ice Sheet occur widely (FAQ 9.1). 35 36 Sea level rise will increase the frequency and severity of extreme sea level events at coasts (see FAQ 8.2),, 37 such as storm surges, wave inundation and tidal floods: risk can be increased by even small changes in 38 global average sea level. Scientists project that in some regions, extreme sea level events that were recently 39 expected once in 100 years will occur annually at 20-25% of locations by 2050 regardless of emissions, but 40 by 2100 emissions choice will matter: annually at 60% of locations for low emissions, and at 80% of 41 locations under strong emissions. 42 43 In many places, local sea level change will be larger or smaller than the global average. From year to year 44 and place to place, changes in ocean circulation and wind can lead to local sea level change. In regions 45 where large ice sheets, such as the Fennoscandian in Eurasia and the Laurentide and Cordilleran in North 46 America, covered the land during the last ice age, the land is still slowly rising up now that the extra weight 47 of the ice sheets is gone. This local recovery is compensating for global sea level rise in these regions and 48 can even lead to local decrease in sea level. In regions just beyond where the former ice sheets reached and 49 the Earth bulged upwards, the land is now falling, and as a result local sea level rise is faster than the global 50 rate. In many regions within low-lying delta regions (such as New Orleans and the Ganges–Brahmaputra 51 delta), the land is rapidly subsiding (sinking) because of human activities such as building dams or 52 groundwater and fossil fuel extraction. Further, when an ice sheet melts it has less gravitational pull on the 53 ocean water nearby. This reduction in gravitational attraction causes sea level to fall close to the (now less- 54 massive) ice sheet while causing sea level to rise farther away. Melt from a polar ice sheet therefore raises 55 sea level most in the opposite hemisphere or in low latitudes – amounting to tens of centimetres difference in Do Not Cite, Quote or Distribute 9-133 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 rise between regions by 2100. 2 3 4 [START FAQ9.2, FIGURE 1 HERE] 5 6 FAQ 9.2, Figure 1: Observed and projected global mean sea level rise and the contributions from its major 7 constituents. 8 9 [END FAQ9.2, FIGURE 1 HERE] 10 11 Do Not Cite, Quote or Distribute 9-134 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 FAQ 9.3: Will the Gulf Stream shut down? 2 3 The Gulf Stream is part of two circulation patterns in the North Atlantic: the Atlantic Meridional 4 Overturning Circulation (AMOC) and the subtropical gyre. Based on models and theory, scientific studies 5 indicate that, while the AMOC is expected to slow in a warming climate, the Gulf Stream will not change 6 much and would not shut down totally, even if the AMOC did. Most climate models project that the AMOC 7 slows in the later 21st century under most emissions scenarios, with some models showing it slowing even 8 sooner. The Gulf Stream affects the weather and sea level, so if it slows, North America will see higher sea 9 levels and Europe’s weather and rate of relative warming will be affected. 10 11 The Gulf Stream is the biggest current in the North Atlantic Ocean. It transports about 30 billion kilograms 12 of water per second northward past points on the east coast of North America. It is a warm current, with 13 temperatures 5°C to 15°C warmer than surrounding waters, so it carries warmer water (thermal energy) from 14 its southern origins and releases warmth to the atmosphere and surrounding water. 15 16 The Gulf Stream is part of two major circulation patterns, the Atlantic Meridional Overturning Circulation 17 (AMOC) and the North Atlantic Subtropical Gyre (FAQ 9.3, Figure 1). The rotation of the Earth causes the 18 big currents in both circulations to stay on the western side of their basin, which in the Atlantic means the 19 circulations combine to form the Gulf Stream. Other large currents contribute to gyres, such as the Kuroshio 20 in the North Pacific and the East Australian Current in the South Pacific, but the Gulf Stream is special in its 21 dual role. There is no comparable deep overturning circulation in the North Pacific to the AMOC, so the 22 Kuroshio plays only one role as part of a gyre. 23 24 The gyres circulate surface waters and result primarily from winds driving the circulation. These winds are 25 not expected to change much and so neither will the gyres, which means the gyre portion of the Gulf Stream 26 and the Kuroshio will continue to transport thermal energy poleward from the equator much as they do now. 27 The gyre contribution to the Gulf Stream is 2 to 10 times larger than the AMOC contribution. 28 29 The Gulf Stream’s role in the AMOC is supplying surface source water that cools, becomes denser and sinks 30 to form cold, deep waters that travel back equatorward, spilling over features on the ocean floor and mixing 31 with other deep Atlantic waters to form a southward current at a depth of about 1500 metres beneath the Gulf 32 Stream. This overturning flow is the AMOC, with the Gulf Stream in the upper kilometre flowing northward 33 and the colder deep water flowing southward. 34 35 The AMOC is expected to slow over the coming centuries. One reason why is freshening of the ocean 36 waters: by meltwater from Greenland, changing Arctic sea ice, and increased precipitation over warmer 37 northern seas. An array of moorings across the Atlantic has been monitoring the AMOC since 2004, with 38 recently expanded capabilities. The monitoring of the AMOC has not been long enough for a trend to emerge 39 from variability and detect long-term changes that may be underway (see FAQ 1.2). Other indirect signs may 40 indicate slowing overturning – for example, slower warming where the Gulf Stream’s surface waters sink. 41 Climate models show that this ‘cold spot’ of slower-than-average warming occurs as the AMOC weakens, 42 and they project that this will continue. Paleoclimate evidence indicates AMOC changed significantly in the 43 past, especially during transitions from colder climates to warmer ones, but indicate it has been stable for 44 8000 years. 45 46 What happens if the AMOC slows in a warming world? The atmosphere adjusts somewhat, compensating 47 partly for the decreases in heat carried by AMOC by carrying more heat. But the ‘cold spot’ makes parts of 48 Europe warm more slowly. Models indicate that weather patterns in Greenland and around the Atlantic will 49 be affected, with reduced precipitation in the mid-latitudes, changing strong precipitation patterns in the 50 tropics and Europe, and stronger storms in the North Atlantic storm track. The slowing of this current 51 combined with the rotation of the Earth means that sea level along North America rises as the AMOC 52 contribution to the Gulf Stream slows. 53 54 The North Atlantic is not the only site of sensitive meridional overturning. Around Antarctica, the world’s 55 densest seawater is formed by freezing into sea ice, leaving behind salty, cold water that sinks to the bottom Do Not Cite, Quote or Distribute 9-135 Total pages: 257 Final Government Distribution Chapter 9 IPCC AR6 WGI 1 and spreads northward. Recent studies show that melting of the Antarctic Ice Sheet and changing winds over 2 the Southern Ocean can affect this southern meridional overturning, affecting regional weather. 3 4 5 [START FAQ9.3, FIGURE 1 HERE] 6 7 FAQ 9.3, Figure 1: Horizontal (gyre) and vertical (Atlantic Meridional Overturning Circulation - AMOC) 8 circulations in the Atlantic today (left) and in a warmer world (right). The Gulf Stream is a 9 warm current composed of both circulations. 10 [END FAQ9.3, FIGURE 1 HERE] 11 12 Do Not Cite, Quote or Distribute 9-136 Total pages: 257