The AR5 assessed from CMIP5 simulations and other lines of evidence that GSAT will continue to rise over the 21st century if greenhouse gas (GHG) concentrations continue increasing (Collins et al., 2013). The AR5 concluded that GSAT for 2081–2100, relative to 1986–2005 will likely be in the 5–95% range of 0.3°C–1.7°C under RCP2.6 and 2.6°C–4.8°C under RCP8.5. The corresponding ranges for the intermediate emissions scenarios with emissions peaking around 2040 (RCP4.5) and 2060 (RCP6.0) are 1.1°C–2.6°C and 1.4°C–3.1°C, respectively. The AR5 further assessed that GSAT averaged over the period 2081–2100 are projected to likely exceed 1.5°C above 1850–1900 for RCP4.5, RCP6.0 and RCP8.5 (high confidence) and are likely to exceed 2°C above 1850–1900 for RCP6.0 and RCP8.5 (high confidence). Global surface temperature changes above 2°C under RCP2.6 were deemed unlikely (medium confidence).

Here, for continuity’s sake, we assess the CMIP6 simulations of GSAT in a fashion similar to the AR5 assessment of the CMIP5 simulations. From these, we compute anomalies relative to 1995–2014 and display the evolution of ensemble means and 5–95% ranges (Figure 4.2). We also use the ensemble mean GSAT difference between 1850–1900 and 1995–2014, 0.82°C, to provide an estimate of the changes since 1850–1900 (Figure 4.2, right axis). Finally, we tabulate the ensemble mean changes between 1995–2014 and 2021–2040, 2041–2060, and 2081–2100 respectively (Figure 4.2).

The CMIP6 models show a 5–95% range of GSAT change for 2081–2100, relative to 1995–2014, of 0.6°C–2.0°C under SSP1-2.6 where CO₂ concentrations peak between 2040 and 2060 (see Table 4.2). The corresponding range under the highest overall emissions scenario (SSP5-8.5) is 2.7°C–5.7°C. The ranges for the intermediate and high emissions scenarios (SSP2-4.5 and SSP3-7.0), where CO₂ concentrations increase to 2100, but less rapidly than SSP5-8.5, are 1.4°C–3.0°C and 2.2°C–4.7°C, respectively. The range for the lowest emissions scenario (SSP1-1.9) is 0.2°C–1.3°C.

In summary, the CMIP6 models show a general tendency toward larger long-term globally averaged surface warming than did the CMIP5 models, for nominally comparable scenarios (very high confidence). In SSP1-2.6 and SSP2-4.5, the 5–95% ranges have remained similar to the ranges in RCP2.6 and RCP4.5, respectively, but the distributions have shifted upward by about 0.3°C (high confidence). For SSP5-8.5 compared to RCP8.5, the 5% bound of the distribution has hardly changed, but the 95% bound and the range have increased by about 20% and 40%, respectively (high confidence). About half of the warming increase has occurred because of more models with higher climate sensitivity in CMIP6, compared to CMIP5; the other half of the warming increase arises from higher effective radiative forcing in nominally comparable scenarios (medium confidence, see Section 4.6.2).

With regards to global warming levels (GWLs) of 1.5°C, 2.0°C and 3.0°C, we note that there is unanimity across all of the CMIP6 model simulations that GSAT change relative to 1850–1900 will rise above: (i) 1.5°C following SSP2-4.5, SSP3-7.0, or SSP5-8.5 (on average around 2030); (ii) 2.0°C following either SSP3-7.0 or SSP5-8.5 (on average around 2043); and (iii) 3.0°C following SSP5-8.5 (on average around 2062). Under SSP1-1.9, 55% and 36% of the model simulations rise above 1.5°C and 2.0°C, respectively, while for SSP1-2.6 those percentages increase to 87% and 58%, respectively. Here, the time of GSAT exceedance is determined as the first year at which 21-year running averages of GSAT exceed the given GWL. In (Section 4.3.4, these values are reassessed using CMIP6 ensemble in combination with other lines of evidence.

CMIP6 models project increases in area-weighted land, ocean, tropical (30°S–30°N), Arctic (67.7°N–90°N), and Antarctic (90°S–55°S) surface air temperature (Table 4.2). Consistent with AR5, and earlier assessments, CMIP6 models project that annual average surface air temperature will warm about 50% more over land than over the ocean, and that the Arctic will warm about more than 2.5 times the global average (Section 4.5.1). For 2081–2100, relative to 1995–2014, the CMIP6 models show 5–95% ranges of warming over land of 0.3°C–2.0°C and 3.5°C–7.6°C following SSP1-1.9 and SSP5-8.5, respectively. The corresponding ranges for Arctic surface air temperature change are 0.5°C–6.6°C and 6.2°C–15.2°C, respectively.

The concentration-driven simulations presented above use a prescribed CO₂ pathway calculated by the MAGICC7.0 model using the CMIP6 emissions (Meinshausen et al., 2020). This is compared here with the CO₂ concentration simulated by CMIP6 ESMs in response to the SSP5-8.5 emissions (Figure 4.3). The 1995–2014 mean simulated CO₂ level is 375 ppm, very similar to the prescribed 378 ppm, but the ESM 5–95% range is 357–391 ppm. By the end of the 21st century (2081–2100), the ESM mean is 953 ppm – below the prescribed CO₂ pathway (1004 ppm), but with a large 5–95% range of 848–1045 ppm, which spans the prescribed concentration level. This result differs from CMIP5, which showed that ESMs typically simulated CO₂ concentrations higher than the prescribed concentration-driven RCP pathways. Reduced spread in CMIP6 carbon cycle feedbacks compared to CMIP5 has been postulated to be due to the inclusion of nitrogen cycle processes in about half of CMIP6 ESMs (Arora et al., 2020). This means that the CMIP6 spread in GSAT response to CO₂ emissions is dominated by climate sensitivity differences between ESMs more than by carbon cycle differences (high confidence) (Jones and Friedlingstein, 2020; Williams et al., 2020).

Figure 4.3 | Comparison ofconcentration-driven and emissions-driven simulation. (a) Atmospheric CO₂ concentration; (b) global surface air temperature from models which performed SSP5-8.5 scenario simulations in both emissions-driven (blue) and concentration-driven (red) configurations. For concentration driven simulations, CO₂ concentration is prescribed, and follows the red line in panel (a) in all models. For emissions-driven simulations, CO₂ concentration is simulated and can therefore differ for each model, blue lines in panel (a). Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Simulated GSAT over 1995–2014, relative to 1850–1900 period, warms by very similar amounts in the two sets of simulations: 0.82°C (0.45–1.31) in emissions-driven compared with 0.75°C (0.53–1.09) in concentration-driven simulations. By the end of the 21st century, warming in emissions-driven simulations is very similar: 4.58°C (3.53–6.70), reflecting the slightly lower CO₂ concentration simulated by the ESMs compared with warming under the prescribed CO₂ pathway of 4.69°C (3.70–6.77). This difference in model-mean response is more than an order of magnitude smaller than the 5–95% spread across model projections. The spread in CO₂ concentration, compared with the prescribed default concentration, leads to a very small increase by about 0.1°C in the spread of GSAT projections, but it is not possible to tell if this is a direct consequence of the simulation configuration or internal variability of the model simulations. These differences due to experimental configuration would be smaller still under scenarios with lower CO₂ levels, and so we assess that results from concentration-driven and emissions-driven configurations do not affect the assessment of GSAT projections (high confidence).

The AR5 assessed from CMIP5 projections that global mean precipitation over the 21st century will increase by more than 0.05 mm day^–1 (about 2% of global precipitation) and 0.15 mm day^–1 (about 5% of global precipitation) under the RCP2.6 and RCP8.5 scenarios, respectively (Collins et al., 2013). These changes are generally in line with those from the CMIP6 simulations following SSP1-2.6 and SSP5-8.5 (Table 4.3).

Figure 4.4 | CMIP6 annual mean precipitation changes (%) from historical and scenario simulations. (a) Northern Hemisphere extratropics (30°N–90°N). (b) North Atlantic subtropics (5°N–30°N, 80°W–0°). Changes are relative to 1995–2014 averages. Displayed are multi-model averages and, in parentheses, 5–95% ranges. The numbers inside each panel are the number of model simulations. Results are derived from concentration-driven simulations. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Unlike AR5, our focus here is on land rather than global precipitation because land precipitation has greater societal relevance. These are displayed as percent changes relative to 1995–2014 (Figure 4.2b). Based on these results, we conclude that global land precipitation is larger during the period 2081–2100 than during the period 1995–2014, under all scenarios considered here (high confidence) (Table 4.3). Global land precipitation for 2081–2100, relative to 1995–2014, shows a 5–95% range of –0.2 to +4.7% under SSP1-1.9 and 0.9–12.9% under SSP5-8.5, respectively. The corresponding ranges under the other emissions scenarios are 0.0–6.6% (SSP1-2.6), 1.5–8.3% (SSP2-4.5), and 0.5–9.6% (SSP3-7.0). A detailed assessment of hydrological sensitivity, or change in precipitation per degree warming, can be found in Chapter 8 (Section 8.2.1).

For scenarios where unanimity across all of the model simulations that GSAT change relative to 1850–1900 rises above 1.5°C (SSP2-4.5, SSP3-7.0, or SSP5-8.5), the ensemble-mean change in global land precipitation from 1850–1900 until the time of exceedance is on average about 1.6%. For scenarios with unanimous global warming above 2.0°C (SSP3-7.0, or SSP5-8.5) and 3.0°C (SSP5-8.5), the ensemble-mean increase in global land precipitation for those models that do exceed 2.0°C and 3.0°C is on average about 2.6% and 4.9%, respectively. On average under SSP1-1.9 and SSP1-2.6, the global land precipitation change for simulations where global warming exceeds 1.5°C and 2.0°C will be about 1.9% and 3.0%, respectively.

Relative to 1995–2014, and across all of the scenarios considered here, CMIP6 models show greater increases in precipitation over land than either globally or over the ocean (Table 4.3; high confidence). Over the Northern Hemisphere (NH) extratropics, the 5–95% changes in precipitation over land between 1995–2014 and 2021–2040, 2041–2060, and 2081–2100, following SSP5-8.5, are 0.6–4.9%, 1.5–8.8%, and 4.7–17.2%, respectively (Figure 4.4). At the other end of scenario spectrum, SSP1-1.9, the corresponding changes are 0.6–5.4%, 0.6–7.3%, and 0.2–7.7%, respectively. By contrast, over the North Atlantic subtropics, precipitation decreases by about 10% following SSP3-7.0 and SSP5-8. There is no change in subtropical precipitation in the North Atlantic following SSP1-1.9, SSP1-2.6, or SSP2-4.5 (high confidence); thereby highlighting the potential limitations of pattern scaling for regional hydrological changes (Section 8.5.3). The reasons for the opposing changes in these two regions are assessed in Chapter 8.

The AR5 assessed from CMIP5 simulations that there will be year-round reductions of Arctic sea ice coverage by the end of this century (Collins et al., 2013). These range from 43% under RCP2.6 and 94% under RCP8.5 in September, and from 8% under RCP2.6 and 34% under RCP8.5 in March (medium confidence). Based on a five-member selection of CMIP5 models, AR5 further assessed that for RCP8.5, Arctic sea ice coverage in September will drop below 1 million km² and be practically ice free at some point between 2040 and 2060. The SROCC further assessed that the probability of an ice-free Arctic in September for stabilized global warming of 1.5°C and 2.0°C is approximately 1% and 10–35%, respectively (IPCC, 2019).

With regards to the model selection in AR5, model evaluation studies have since identified shortcomings of the CMIP5 models to match the observed distribution of sea ice thickness in the Arctic (Stroeve et al., 2014; Shu et al., 2015) and the observed evolution of albedo on seasonal scales (Koenigk et al., 2014). It was also found that many models’ deviation from observed sea ice cover climatology cannot be explained by internal variability, whereas the models’ deviation from observed sea ice cover trend (over the satellite period) can often be explained by internal variability (Olonscheck and Notz, 2017). This hinders a selection of models according to their simulated trends, which additionally has been shown to only have a weak effect on the magnitude of simulated future trends (Stroeve and Notz, 2015).

Based on results from the CMIP6 models, we conclude that on average the Arctic will become practically ice-free in September by the end of the 21st century under SSP2-4.5, SSP3-7.0, and SSP5-8.5 (high confidence) (Figure 4.2c and Table 4.4). Also, in the CMIP6 models, Arctic SIA in March decreases in the future, but to a much lesser degree, in percentage terms, than in September (high confidence) (Table 4.4). A more detailed assessment of projected Arctic and also Antarctic sea ice change can be obtained in Chapter 9 (Section 9.3.1).

Studies focusing on the relationship of sea ice extent and changes in external drivers have consistently found a much-reduced likelihood of a practically ice-free Arctic Ocean during summer for global warming of 1.5°C than for 2.0°C (Screen and Williamson, 2017; Jahn, 2018; Niederdrenk and Notz, 2018; Notz and Stroeve, 2018; Sigmond et al., 2018; Olson et al., 2019). This is shown here in a large initial-condition ensemble of observationally constrained model simulations where GSAT are stabilized at 1.5°C, 2.0°C and 3.0°C warming relative to 1850–1900 in the RCP8.5 scenario (Figure 4.5). Temperature stabilization is achieved by switching off all the anthropogenic emissions around the time that GSAT first reaches the stabilization thresholds. Simulations have been observationally constrained to correct for a model bias in simulated historical September sea ice extent. In these simulations, Arctic sea ice coverage in September is simulated, on average, to drop below 1 million km² around 2040, consistent with the AR5 set of assessed models (Sigmond et al., 2018). The individual model simulations, for which there are twenty for each stabilized temperature level, show that the probability of the Arctic becoming practically ice free at the end of the 21st century is significantly higher for 2°C warming than for 1.5°C warming above 1850–1900 levels (high confidence).

Figure 4.5 | Arctic sea ice extent in September in a large initial-condition ensemble of observationally-constrained simulations of an Earth system model (CanESM2). The black and red curves are averages over twenty simulations following historical forcings to 2015 and RCP8.5 extensions to 2100. The other curves are averages of over 20 simulations each after global surface air temperature has been stabilized at the indicated degree of global mean warming relative to 1850–1900. The bars to the right are the minimum to maximum ranges over 2081–2100 (Sigmond et al., 2018). The horizontal dashed line indicates a practically sea ice-free Arctic. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

The AR5 assessed from CMIP5 process-based simulations that the rate of GMSL rise during the 21st century will very likely exceed the rate observed during 1971–2010 for all RCP scenarios due to increases in ocean warming and loss of mass from glaciers and ice sheets (Church et al., 2013). Further, AR5 concluded that for the period 2081–2100, compared to 1986–2005, GMSL rise is likely (medium confidence) to be in the 5–95% range of projections from process-based models, which give 0.26–0.55 m for RCP2.6, 0.32–0.63 m for RCP4.5, 0.33–0.63 m for RCP6.0, and 0.45–0.82 m for RCP8.5. For RCP8.5, the rise by 2100 is 0.52–0.98 m with a rate during 2081–2100 of 8–16 mm yr^–1.

There have been substantial modelling advances since AR5, with most sea level projections corresponding to one of three categories: (i) central-range projections, combining scenario-conditional probability distributions for the different contributions to estimate a central range under different scenarios; (ii) probabilistic projections, which explicitly consider outcomes for a wide range of likelihoods, including low-likelihood, high-impact outcomes; and (iii) semi-empirical projections, based on statistical relationships between past GMSL changes and climate variables, which now calibrate individual contributions and are consistent with physical-model based estimates (Section 9.6.3).

Based on the assessment of the latest modelling information (Figure 4.2d and Section 9.6.3), we conclude that under the SSP3-7.0, the likely range of GMSL change averaged over 2081–2100 relative to 1995–2014 is 0.46–0.74 m. Under SSP1-2.6, the likely range over the long-term is 0.30–0.54 m. Further, in SSP2-4.5, SSP3-7.0, and SSP5-8.5, the rise in GMSL is projected to accelerate over the 21st century. A detailed assessment of the processes contributing to these projected rises and accelerations in GMSL, together with a comparison to AR5 and SROCC, can be found in Chapter 9 (Section 9.6.3). Projected changes in the thermosteric component of GMSL beyond 2300 are assessed in Section 4.7.1.

In summary, it is virtually certain that under any one of the assessed SSPs, there will be continued rise in GMSL through the 21st century.

The AR5 assessed from CMIP5 simulations that the Atlantic Meridional Overturning Circulation (AMOC) will very likely weaken over the 21st century, and the projected weakening of the AMOC is consistent with CMIP5 projections of an increase of high-latitude temperature and high-latitude precipitation, with both effects causing the surface waters at high latitudes to become less dense and therefore more stable (Collins et al., 2013).

Based on CMIP6 models, we find that over the 21st century, AMOC strength, relative to 1995–2014, shows a multi-model mean decrease in each of the SSP scenarios but with a large spread across the individual simulations (Figure 4.6). We also note that the magnitude of the ensemble-mean strength decrease is approximately scenario independent up to about 2060 (Weijer et al., 2020). A more detailed assessment of these projected AMOC changes, and the mechanisms involved, can be found in Chapter 9 (Section 9.2.3).

Figure 4.6 | CMIP6 annual mean Atlantic Meridional Overturning Circulation (AMOC) strength change in historical and scenario simulations. Changes are relative to averages from 1995–2014. The curves show ensemble averages and the shadings the 5–95% ranges across the SSP1-2.6 and SSP3-7.0 ensembles. The circles to the right of the panel show the anomalies averaged from 2081–2100 for each of the available model simulations. The numbers inside the panel are the number of model simulations. Here, the strength of the AMOC is computed as the maximum value of annual mean ocean meridional overturning mass stream function in the Atlantic at 26°N. Results are from concentration-driven simulations. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

In summary, we assess from the CMIP6 models that AMOC weakening over the 21st century is very likely ; the rate of weakening is approximately independent of the emissions scenario (high confidence).

Based on a large initial condition ensemble of simulations with a CMIP5 model (CanESM2) with emissions scenarios leading to stabilization of global warming of 1.5°C, 2.0°C, or 3.0°C relative to 1850–1900, AMOC continues to decline for 5–10 years after GSAT is effectively stabilized at the given GWL (Sigmond et al., 2020). This is followed by a recovery of AMOC strength for about the next 150 years to a level that is approximately independent of the considered stabilization scenario. These results are replicated in simulations in a CMIP6 model (CanESM5) with emissions cessation after diagnosed CO₂ emissions reach 750 Gt, 1000 Gt, or 1500 Gt. These emissions levels lead to global warming stabilization at 1.5°C, 2.0°C, or 3.0°C relative to 1850–1900. In summary, in these model simulations the AMOC recovers over several centuries after the cessation of CO₂ emissions (medium confidence).

The AR5 concluded with very high confidence that ocean carbon uptake of anthropogenic CO₂ will continue under all RCPs through the 21st century, with higher uptake corresponding to higher concentration pathways. The future evolution of the land carbon uptake was assessed to be much more uncertain than for ocean carbon uptake, with a majority of CMIP5 models projecting a continued cumulative carbon uptake.

Based on results from the CMIP6 models, we conclude that the flux of carbon from the atmosphere into the ocean increases continually through most of 21st century in the two highest emissions and decreases continually under the other emissions scenarios (Figure 4.7a). The flux of carbon from the atmosphere to land shows a similar 21st century behaviour across the scenarios but with much higher year-to-year variation than ocean carbon flux (Figure 4.7b). A more in-depth assessment and discussion of the mechanism involved can be found in Chapter 5 (Section 5.4.5).

Figure 4.7 | CMIP6 carbon uptake in historical and scenario simulations. (a) Atmosphere to ocean carbon flux (PgC yr^–1). (b) Atmosphere to land carbon flux (PgC yr^–1). The curves show ensemble averages and the shadings show the 5–95% ranges across the SSP1-2.6 and SSP3-7.0 ensembles. The numbers inside each panel are the number of model simulations. The land uptake is taken as Net Biome Productivity (NBP) and so includes any modelled net land-use change emissions. Results are from concentration-driven simulations. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

In summary, we assess that the cumulative uptake of carbon by the ocean and by land will increase through the 21st century irrespective of the considered emissions scenarios except SSP1-1.9 (very high confidence).

The AR5 assessed from CMIP5 simulations that it is virtually certain that increasing storage of carbon by the ocean under all four RCPs through to 2100 will increase ocean acidification in the future (Ciais et al., 2013). Specifically, AR5 reported that CMIP5 models project increased ocean acidification globally to 2100 under all RCPs, and that the corresponding model mean and model spread in the decrease in surface ocean pH from 1986–2005 to 2081–2100 would be 0.065 (0.06–0.07) for RCP2.6, 0.145 (0.14–0.15) for RCP4.5, 0.203 (0.20–0.21) for RCP6.0 and 0.31 (0.30–0.32) for RCP8.5.

Based on results from the CMIP6 models we conclude that, except for the lower-emissions scenarios SSP1-1.9 and SSP1-2.6, ocean surface pH decreases monotonically through the 21st century (high confidence) (Figure 4.8).

Figure 4.8 | Global average surface ocean pH. The shadings around the SSP1-2.6 and SSP5-7.0 curves are the 5–95% ranges across those ensembles. The numbers inside each panel are the number of model simulations. Results are from concentration-driven simulations. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

The El Niño–Southern Oscillation (ENSO) is the most dominant mode of variability on interannual time scales and also the dominant source of seasonal climate predictability (Box 11.3 and Annex IV, Section AIV.2.3; Timmermann et al., 2018). The AR5 assessed from CMIP5 simulations that ENSO variability will very likely remain the dominant mode of interannual climate variability in the future, and that associated ENSO precipitation variability on regional scales is likely to intensify (Christensen et al., 2013). However, they assessed there was low confidence in projected changes in ENSO variability in the 21st century due to a strong component of internal variability.

Among a range of indices proposed for representing ENSO, we use the most prominent one, the Niño 3.4 index, defined as the average equatorial SST or precipitation across the central equatorial Pacific (5°S–5°N, 170°W–120°W; Section AIV.2.3). Here, we consider the evolution of the amplitude of Niño 3.4 index for SST and precipitation over the 21st century as projected by CMIP6 models. Analysis of CMIP6 models shows there is no robust model consensus on the forced changes in the amplitude of ENSO SST variability even under the high-emissions scenarios SSP3-7.0 and SSP5-8.5, but a significant increasing trend in the amplitude of ENSO precipitation variability is projected across the 21st century in the four SSPs (Figure 4.10). This is broadly consistent with results from CMIP5 models (Christensen et al., 2013; Power et al., 2013; Cai et al., 2015; Chen et al., 2017; Wengel et al., 2018), recent studies with CMIP6 models (Brown et al., 2020; Fredriksen et al., 2020; Freund et al., 2020; Yun et al., 2021), and large initial-condition ensemble experiments (Maher et al., 2018; Zheng et al., 2018; Haszpra et al., 2020).

Figure 4.10 | Changes in amplitude of ENSO Variability. Variability of (a) SST and (b) precipitation anomalies averaged over Niño 3.4 region for 1950–2014 from CMIP6 historical simulations and for 2015–2100 from four SSPs. Thick lines stand for multi-model mean and shading is the 5–95% range across CMIP6 models for historical simulation (grey), SSP1-2.6 (blue) and SSP3-7.0 (pink), respectively. The amplitude of ENSO SST and rainfall variability is defined as the standard deviation of the detrended Niño 3.4-area averaged SST and rainfall index, respectively, over 30-year running windows. The standard deviation in every single model is normalized by each model’s present-day standard deviation averaged from 1995 to 2014. The number of available models is listed in parentheses. This figure is adopted from Yun et al. (2021). Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

It is therefore very likely that the amplitude of ENSO rainfall variability will intensify in response to global warming over the 21st century although there is no robust consensus from CMIP6 climate models for a systematic change in amplitude of ENSO SST variability even in the high-emissions scenarios of SSP3-7.0 and SSP5-8.5.

The AR5 assessed that it is likely that GSAT will increase in the range 0.3°C–0.7°C over the period 2016–2035 relative to 1986–2005 (medium confidence), and that there were not large differences in the GSAT change among different RCPs in this period (Kirtman et al., 2013). The AR5 further assessed that it is more likely than not that the mean GSAT for the period 2016–2035 will be more than 1°C above the mean for 1850–1900, and it is very unlikely that it will be more than 1.5°C above the 1850–1900 mean (medium confidence). It was shown that in the period 2016–2035, differences in GSAT across RCP scenarios for a single climate model are typically smaller than differences between climate models under a single RCP scenario, indicating that model structural uncertainty is larger than scenario uncertainty over that period (Hawkins and Sutton, 2009).

Near-term (2021–2040) GSAT changes relative to 1995–2014 exhibit only minor dependence on SSP scenario, consistent with AR5 (Table 4.5). Averaged over the twenty years of the near term and across all scenarios, GSAT is very likely to be higher than over 1995–2014 by 0.4°C–1.0°C (Table 4.5), with most of the uncertainty arising from that in ECS and TCR (high confidence) (Section 4.3.4; e.g., Lehner et al., 2020). The assessed near-term warming is thus larger than in AR5 by 0.1°C to 0.2°C. This upward revision has the same magnitude as the ad-hoc downward adjustment to near-term projected GSAT change that was performed in AR5 (Box 4.1; Kirtman et al., 2013).

Averaged near-term GSAT is as likely as not at least 1.5°C higher than during 1850–1900, across the five SSP scenarios used here (Table 4.5 and Section 4.3.4). This much higher likelihood of near-term warming reaching 1.5°C than in AR5 arises both because surface warming has continued since AR5 (the period 1995–2014 was warmer by 0.16°C than 1986–2005; Cross-Chapter Box 2.3, Table 1), and because of methodological and dataset updates (the AR6 assessment of 1986–2005 GSAT change relative to 1850–1900 is 0.08°C higher than in the AR5; Cross-Chapter Box 2.3, Table 1).

For annual mean GSAT, uncertainty in near-term projections arises in roughly equal measure from internal variability and model uncertainty (high confidence) (Box 4.1). Forecasts initialized from recent observations simulate GSAT changes for the period 2019–2028 relative to the recent past that are consistent with the assessed very likely range in annual mean GSAT (high confidence) (Box 4.1, Figure 1, and Table 4.5). Because annual mean GSAT shows a higher level of internal variability than the 20-year mean, individual years are expected to cross the 1.5°C earlier than the assessed GSAT does. For example, Smith et al. (2018) apply a multi-model decadal-forecast ensemble to assess the likelihood that global warming of 1.5°C higher than over 1850–1900 will be temporarily exceeded in the near future.

When we repeat the uncertainty quantification for GSAT as in Section 4.3.4 but with the corresponding higher level of internal variability for annual instead of 20-year averages added in quadrature, we can estimate the likelihood that an individual year would cross the GSAT 1.5°C threshold. By 2030, GSAT in any individual year could exceed 1.5°C relative to 1850–1900 with a likelihood between 40 and 60 percent, across the scenarios considered here (medium confidence).

Consistent with AR5 and earlier assessments, Figure 4.12 shows for SSP1-2.6 and SSP3-7.0 that the largest warming occurs at high latitudes, particularly in boreal winter in the Arctic (Section 4.5.1.1), and larger warming over land than over the ocean (Section 4.5.1.1). In both scenarios, the increase in seasonal mean surface temperatures over many NH land regions exceeds 1°C relative to 1995–2014. In the near term, the two scenarios show surface temperature changes that are similar in magnitude. The trajectories for well-mixed GHGs, and as a consequence the effective radiative forcing, in the scenarios have not yet diverged that much (O’Neill et al., 2016; Riahi et al., 2017). Based on the currently available CMIP6 models, regions that do not show robust warming in the near-term include the northern North Atlantic, parts of India, parts of North America and Eurasia in winter, and the subtropical eastern Pacific in the Southern Hemisphere.

Figure 4.12 | Near-term change of seasonal mean surface temperature. Displayed are projected spatial patterns of CMIP6 multi-model mean change (°C) in (top) December–January–February (DJF) and (bottom) June–July–August (JJA) near-surface air temperature for 2021–2040 from SSP1-2.6 and SSP3-7.0 relative to 1995–2014. The number of models used is indicated in the top right of the maps. No overlay indicates regions where the change is robust and likely emerges from internal variability, that is, where at least 66% of the models show a change greater than the internal-variability threshold (Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

The ERF patterns from aerosols and well-mixed GHGs are distinct (Chapter 7), and warming patterns therefore depend on the precise mix of forcing agents in the scenarios. The spatial efficacies – the change in surface temperature per unit ERF – for CO₂, sulphate and black carbon aerosols and solar forcing have been recently evaluated in climate models (Modak et al., 2016, 2018; Duan et al., 2018; Modak and Bala, 2019; Richardson et al., 2019). On average, the spatial patterns of near-surface warming are largely similar for different external drivers (Xie et al., 2013; Richardson et al., 2019; Samset et al., 2020), despite the patterns of forcing being different and despite the large spread across different models (Richardson et al., 2019).

Internal variability in near-surface temperature change is large in many regions, particularly in mid-latitudes and polar regions (Hawkins and Sutton, 2012). Projections from individual realizations can therefore exhibit divergent regional responses in the near-term in areas where the amplitude of a forced signal is relatively small compared to internal variability (Deser et al., 2012b, 2014, 2016).

The AR5 assessed that zonal mean precipitation will very likely increase in high and some of the mid latitudes and will more likely than not decrease in the subtropics. The AR5 further assessed that the near-term changes in precipitation are largely uncertain at regional scales, and much of the non-robustness in near-term projections is attributable to internal variability and model uncertainty.

The mean patterns of seasonal precipitation change in CMIP6 models are consistent with AR5, increasing at high latitudes, over oceanic regions, and in wet regions over the tropics; and decreasing in dry regions including large parts of the subtropics (Figure 4.13). The magnitude of projected changes in precipitation in the near term, especially on regional scales is small compared to the magnitude of internal variability (Section 10.4.3; Hawkins and Sutton, 2011, 2016; Hoerling et al., 2011; Deser et al., 2012b; Power et al., 2012). Analyses of CMIP5, CMIP6, and single-model large-ensemble simulations show that for the uncertainty in near-term precipitation projections, model uncertainty and internal variability dominate while the scenario uncertainty is very small (Section 8.5; Lehner et al., 2020). Based on large ensembles of climate change experiments, it was shown that internal variability decreases over time for both temperature and precipitation on decadal scales (Zhang and Delworth, 2018; Tebaldi et al., 2021). The precipitation projections from CMIP6 models shows larger model uncertainty associated with the higher average transient climate response (Lehner et al., 2020).

Figure 4.13 | Near-term change ofseasonal mean precipitation. Displayed are projected spatial patterns of CMIP6 multi-model mean change (%) in (top) December–January–February (DJF) and (bottom) June–July–August (JJA) precipitation from SSP1-2.6 and SSP3-7.0 in 2021–2040 relative to 1995–2014. The number of models used is indicated in the top right of the maps. No overlay indicates regions where the change is robust and likely emerges from internal variability, that is, where at least 66% of the models show a change greater than the internal-variability threshold (Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

The ‘wet get wetter, dry get drier’ paradigm, which has been used to explain the global precipitation pattern responding to global warming (Held and Soden, 2006), might not hold, especially over subtropical land regions (Greve et al., 2014; Feng and Zhang, 2015; Greve and Seneviratne, 2015). Over the tropical oceans, precipitation changes are largely driven by the pattern of SST changes (He et al., 2018), and in the subtropics, precipitation response is driven primarily by the fast adjustment to CO₂ forcing (He and Soden, 2017). In addition to the response to GHG forcing, forcing from natural and anthropogenic aerosols exert impacts on regional patterns of precipitation (Section 10.3.1; Ramanathan et al., 2005; Bollasina et al., 2011; Polson et al., 2014; Krishnan et al., 2016; L. Liu et al., 2018; Shawki et al., 2018). The large uncertainties in near-term regional precipitation projections arise due to the interplay between internal variability and anthropogenic external forcing (Endo et al., 2018; Wang et al.,2021). Uncertainties in future aerosol emissions scenarios contribute to uncertainties in regional precipitation projections (Wilcox et al., 2020). Aerosol changes induce a drying in the SH tropical band compensated by wetter conditions in the NH counterpart (Acosta Navarro et al., 2017). The spatially uneven distribution of the aerosol forcing may also induce changes in tropical precipitation caused by shifts in the mean location of the intertropical convergence zone (ITCZ) (Hwang et al., 2013; Ridley et al., 2015; Voigt et al., 2017). Because of the large uncertainty in the aerosol radiative forcing and the dynamical response to the aerosol forcing there is medium confidence in the impacts of aerosols on near-term projected changes in precipitation. Precipitation changes in the near term show seasonal amplification, precipitation increase in the rainy season and decrease in the dry season (Fujita et al., 2019).

Consistent with AR5, we conclude that projected changes of seasonal mean precipitation in the near term will increase at high latitudes. Near-term projected changes in precipitation are uncertain mainly because of natural internal variability, model uncertainty, and uncertainty in natural and anthropogenic aerosol forcing (medium confidence).

The global monsoon is a forced response of the coupled atmosphere–land–ocean system to the annual cycle of solar insolation and characterized by a seasonal reversal of circulation and a seasonal alternation of dry and wet conditions (Section 8.3.2, Figure 8.11 and Annex V). The global monsoon concept helps to dissect the mechanisms and controlling factors of monsoon variability at various temporal-spatial scales (Wang and Ding, 2008; P.X. Wang et al., 2017).

In AR5, there was no specific assessment on global monsoon changes in the near term, but information can be derived from CMIP5 projections of the spatial patterns of precipitation change. While the basic pattern of wet regions, including global monsoon regions, tending to get wetter and dry regions tending to get drier is apparent, large response uncertainty is evident in the substantial spread in the magnitude of projected change from different simulations. Over the global land monsoon regions, model uncertainty and internal variability together explain 99.7% of the fraction of total variance (Zhou et al., 2020), near-term projected multi-model mean precipitation changes are almost everywhere smaller than the estimated standard deviation of internal variability (Figure 4.13).

The global land monsoon precipitation index, defined as the area-weighted precipitation rate in the global land monsoon domain, tends to increase in the near term under all five core SSPs (Figure 4.14a) (Chen et al., 2020), but changes are small compared to the intermodel spread in the historical period. The Northern Hemisphere summer monsoon circulation index, defined as the vertical shear of zonal winds between 850 and 200 hPa averaged in a zone stretching from Mexico eastward to the Philippines (0°–20°N, 120°W–120°E), tends to decrease under four of the five SSP scenarios (Figure 4.14b), potentially offsetting monsoon precipitation increase. Projected changes in the global monsoon circulation are also uncertain, because they are influenced by internal variability such as AMV and PDV (see Section 3.3.3.2) and structural differences across models. In the near-term, for CMIP6 projections (Figure 4.14a), the multi-model mean (5–95% range) of global land monsoon precipitation change is 1.9% (–0.4 to 4.9%), 1.6% (–1.0 to 5.2%), 1.3% (–1.7 to 3.7%), and 1.9% (–0.8 to 5.2%) under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, respectively.

Figure 4.14 | Time series of global land monsoon precipitation and Northern Hemisphere summer monsoon (NHSM) circulation index anomalies. (a) Global land monsoon precipitation index anomalies (unit: %) defined as the area-weighted mean precipitation rate in the global land monsoon domain (as defined by Wang et al. (2013a) for the CMIP6 historical simulation (1950–2014) and five SSPs (2015–2100). (b) Anomalies in NHSM circulation index (unit: m s^–1), defined as the vertical shear of zonal winds between 850 and 200 hPa averaged in a zone stretching from Mexico eastward to the Philippines (0°–20°N, 120°W–120°E; Wang et al., 2013a) for the CMIP6 historical simulation and five SSPs. One realization is averaged from each model. Anomalies are shown relative to the present-day (1995–2014) mean. The curves show averages over the simulations, the shadings around the SSP1-2.6 and SSP3-7.0 curves show 5–95% ranges, and the numbers near the top show the number of model simulations used. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

In summary, we assess that near-term changes in global monsoon precipitation and circulation will be affected by the combined effects of model uncertainty and internal variability, such as AMV and PDV, which together are larger than the forced signal (medium confidence).

The AR5 assessed that for RCP8.5, Arctic sea ice coverage in September will drop below 1 million km², or become practically ice-free, at some point between 2040 and 2060 (Collins et al., 2013). Since AR5, there has been substantial progress in understanding the response of Arctic sea ice to near-term changes in external forcing. In particular, it is very likely that different trajectories of the near-term evolution of anthropogenic forcing cause distinctly different likelihood ranges for very low sea ice coverage to occur over the next two decades (Notz and Stroeve, 2018). For example, there is an unlikely drop of September Arctic sea ice coverage to below 1 million km² before 2040 for RCP 2.6, and a likely drop of September Arctic sea ice coverage to below 1 million km² before 2040 for RCP 8.5 (medium confidence given the single study). The much higher likelihood of a practically sea ice free Arctic Ocean during summer before 2040 in RCP8.5 compared to RCP2.6 is consistent with related studies assessed in SROCC that find a substantially increased likelihood of an ice-free Arctic Ocean for 2.0°C compared to 1.5°C mean global warming relative to pre-industrial levels (Screen and Williamson, 2017; Jahn, 2018; Niederdrenk and Notz, 2018; Notz and Stroeve, 2018; Sigmond et al., 2018; Olson et al., 2019).

Based on results from CMIP6 models, we conclude that Arctic SIA will decrease in September in the near term (Figure 4.15, high confidence). In the case of 10-year trends ending in the near term, 79% of the simulations considered across all the core SSPs project decreasing Arctic sea ice area in September. Due to less of an influence from internal variability, this number rises to 98% in the case of 30-year trends. A more detailed assessment of near-term Arctic sea ice changes can be found in Chapter 9 (Section 9.3.1). A detailed assessment of Antarctic sea ice changes is in Chapter 9 (Section 9.3.2).

Figure 4.15 | CMIP6 linear trends in September Arctic sea-ice area for 10-year, 20-year, and 30-year periods ending in 2021–2040 following five SSPs. Plotted are the 5–95% ranges across the ensembles of simulations. The numbers at the top of the plot are the number of model simulations in each SSP ensemble. The numbers near the bottom of the plot indicate the percentage of simulations across all the SSPs with decreasing sea-ice area. Results are from concentration-driven simulations. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Ocean carbon flux is both a key feature of the physical ocean in mitigating the rise of atmospheric CO₂ and a driver of changes in the ocean biosphere, including changes in ocean acidity. Based on results from CMIP6 models, we conclude that SSP2-4.5, SSP3-7.0, and SSP5-8.5 all clearly lead to increasing 10-, 20-, and 30-year trends in ocean carbon flux over the near term (high confidence) (Figure 4.16,). Increasing trends in ocean carbon flux are less obvious in the lower-emissions scenarios. Ensemble-mean trends in land carbon flux over the near term are generally increasing, but these are unlikely to be detected given a large component of terrestrial variability combined with model uncertainty. A more detailed assessment is in Chapter 5 (Section 5.2.1).

Figure 4.16 | CMIP6 trends in ocean and land carbon flux for 10-year, 20-year, and 30-year periods ending in 2021–2040. (a) Ocean carbon flux. (b) Land carbon flux. Plotted are the 5–95% ranges across the ensembles of simulations, for five SSPs. The numbers at the top of the plots are the number of model simulations in each SSP ensemble. Units are PgC yr^–1 per decade. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

In summary, it is likely that ocean carbon flux will increase in the near term under the higher emissions scenarios, while a large component of terrestrial variability makes it is unlikely that an increase in land carbon flux will be detected over this period.

The AR5 assessed that it is very likely that the ENSO will remain the dominant mode of interannual variability in the future but did not specify its change in near term. A subset of CMIP5 models that simulate the ENSO Bjerknes index most realistically show an increase of ENSO SST amplitude in the near-term future and decline thereafter (Kim et al., 2014). However, detection of robust near-term changes of ENSO SST variability in response to anthropogenic forcing is difficult to achieve due to pronounced unforced low-frequency modulations of ENSO (Wittenberg, 2009; Maher et al., 2018; Wengel et al., 2018). Figure 4.10 in Section 4.3.3.2, using CMIP6 models, also shows no robust change in ENSO SST variability in the near term.

While there is no strong model consensus on the change in amplitude of ENSO SST variability, the amplitude of ENSO-associated rainfall variabilitylikely increases (Power et al., 2013; Cai et al., 2015). Analysis of CMIP6 models shows a slight increasing trend in amplitude of rainfall variability over Niño 3.4 region in the near term attributable to mean moisture increase, regardless of changes in ENSO SST variability (Figure 4.10). However, there are no distinguishable changes in the rainfall variability among five SSPs with significant model spread in the near term. Hence, no robust change in amplitude of ENSO SST and rainfall variability is expected in the near term although the rainfall variability slightly increases (medium confidence).

Important modes of interannual climate variability with pronounced climate impacts in the Africa–Indo-Pacific areas of the globe are the Indian Ocean Dipole (IOD), which is closely related to, and often coincides with, ENSO phases (Stuecker et al., 2017), and the Indian Ocean basin (IOB) mode. This is often described as a capacitor effect in response to ENSO (Xie et al., 2009; Du et al., 2013) and can feed back onto ENSO evolution (Cai et al., 2019). IOD and IOB are extensively described in Annex IV (Section AIV2.4).

The projected climate mean state changes in the tropical Indian Ocean resemble a positive IOD state, with faster warming in the west compared to the east. This mean state change will potentially lead to a reduction in the amplitude of IOD events, albeit with no robust change in IOD frequency (Cai et al., 2014b). There is no robust evidence yet suggesting a cessation of IOD variability or a significant change in the IOB mode in the near-term.

Interannual variability of the tropical Atlantic can be described in terms of two main climate modes: the Atlantic equatorial mode and the Atlantic meridional mode (AMM; Annex IV, Section AIV2.5). The Atlantic equatorial mode, also commonly referred to as the Atlantic Niño or Atlantic Zonal Mode, is associated with SST anomalies near the equator, peaking in the eastern basin, while the AMM is characterized by an inter-hemispheric gradient of SST and wind anomalies. Both modes are associated with changes in the ITCZ and related winds and exert a strong influence on the climate in adjacent and remote regions.

Despite considerable improvements in CMIP5 with respect to CMIP3, most CMIP5 models have difficulties in simulating the mean climate of the tropical Atlantic (Mohino et al., 2019) and are not able to correctly simulate the main aspects of Tropical Atlantic Variability (TAV) and associated impacts. This is presumably the main reason why there is a lack of specific studies dealing with near-term changes in tropical Atlantic modes. Nevertheless, AR5 reported that the ocean is more predictable than continental areas at the decadal time scale (Kirtman et al., 2013). In particular, the predictability in the tropical oceans is mainly associated with decadal variations of the external forcing component. Since the AMV affects the tropical Atlantic, near-term variations of the AMV can modulate the equatorial mode and the AMM as well as associated impacts.

There are no specific studies focusing on near-term changes in tropical Atlantic modes; nevertheless, decadal predictions show that although the North Atlantic stands out in most CMIP5 models as the primary region where skill might be improved because of initialization, encouraging results have also been found in the tropical Atlantic (Meehl et al., 2014). The effect of initialization in the tropical Atlantic is not only visible in surface temperature but also in the subsurface ocean (Corti et al., 2015). In particular, initialization improves the skill via remote ocean conditions in the North Atlantic subpolar gyre and tropical Pacific, which influence the tropical Atlantic through atmospheric teleconnections (Dunstone et al., 2011; Vecchi et al., 2014; García-Serrano et al., 2015). Improvements of some aspects of climate prediction systems (initialization techniques, large ensembles, increasing model resolution) have also led to skill improvements over the tropical Atlantic (Pohlmann et al., 2013; Monerie et al., 2017; Yeager and Robson, 2017).

Recent studies have shown that the AMV can modulate not only the characteristics of the Atlantic Niños, but also their inter-basin teleconnections (Indian and Pacific). In particular, the Atlantic Niño–ENSO relationship is strongest during negative AMV phases (Martín-Rey et al., 2014; Losada and Rodríguez-Fonseca, 2016) when equatorial Atlantic SST variability is enhanced (Martín-Rey et al., 2017; Lübbecke et al., 2018).

Based on CMIP5 and available CMIP6 results, we conclude that there is a lack of studies on the near-term evolution of TAV and associated teleconnections for a comprehensive assessment. However, some studies show that despite severe model biases there are skilful predictions in the mean state of tropical Atlantic surface temperature several years ahead (medium confidence), though skill in simulated variability has not been assessed yet. Decadal changes in the Atlantic Niño spatial configuration and associated teleconnections might be modulated by the AMV, but there is limited evidence and therefore low confidence in these results.

Climate variability of the Pacific Ocean on decadal and inter-decadal time scales is described in terms of a number of quasi-oscillatory SST patterns such as the Pacific Decadal Oscillation (PDO; Mantua et al., 1997) and the Inter-decadal Pacific Oscillation (IPO; Folland, 2002), which are referred to as the Pacific Decadal Variability (PDV; Newman et al., 2016). PDV comprises an inter-hemispheric pattern that varies at decadal to inter-decadal time scales (Figure 3.35). However, although the spatial domains to derive the IPO and PDO indices differ, and despite uncertainty related to trend removal and time-filtering (Newman et al., 2016; Tung et al., 2019), the IPO and PDO are highly correlated in time and they will be assessed together as the PDV (Annex IV, Section AIV.2.6).

The AR5 assessed that near-term predictions of PDV (then referred to as PDO or IPO) were largely model dependent (Mochizuki et al., 2012; van Oldenborgh et al., 2012), not robust to sampling of initialization start-dates, overall not statistically significant, and worse than persistence (Doblas-Reyes et al., 2013), although some studies showed positive skill for PDV (Mochizuki et al., 2010; Chikamoto et al., 2013). The CMIP5 decadal-prediction ensemble yielded no prediction skill of SST over the key PDV centres of action in the Pacific Ocean, both at two-to-five-year and six-to-nine-year forecast averages (Doblas-Reyes et al., 2013; Guemas et al., 2013; Boer and Sospedra-Alfonso, 2019).

Since AR5, the processes causing the multi-decadal variability in the Pacific Ocean have become better understood (Newman et al., 2016; Henley, 2017). However, the relative importance oftropical and extratropical processes underlying PDV remains unclear; although it seems to be stochastically driven rather than self-excited (Liu, 2012; Liu and Di Lorenzo, 2018), the South Pacific being a key region for the tropical branch of PDV (Chung et al., 2019; Liguori and Di Lorenzo, 2019).

Because PDV represents not one, but many dynamical processes, it represents a challenge as a target for near-term climate predictions and projections. The new generation of decadal forecast systems keeps showing poor (Shaffrey et al., 2017) to moderate (D.M. Smith et al., 2019) multi-year prediction skill for PDV, although the potential for forecasting capabilities is demonstrated in case studies (Meehl and Teng, 2012; Meehl et al., 2014). For the near-term, a transition of PDV from the negative phase (1999–2012) towards a positive phase is predicted in the coming years (2013–2022; Meehl et al., 2016).

The PDV has been shown to influence the pace of global warming (Cross-Chapter Box 3.1), but the extent to which PDV is externally forced or internally generated (Mann et al., 2020) remains an open question, and there is still no robust evidence. Thus, there is low confidence on how the PDV will evolve in the near-term (Bordbar et al., 2019).

The Atlantic Multi-decadal Variability (AMV) is a large-scale climate mode accounting for the main fluctuations in North Atlantic SST on multi-decadal time scales (Section AIV.2.7). The AMV influences air temperatures and precipitation over adjacent and remote continents, and its undulations can partially explain the observed variations in the NH mean temperatures (Steinman et al., 2015). The origin of this variability is still uncertain. Ocean dynamics (e.g., changes in the AMOC), external forcing, and local atmospheric forcing all seem to play a role (Menary et al., 2015; Ruprich-Robert and Cassou, 2015; Brown et al., 2016; Cassou et al., 2018; Wills et al., 2019). Recent studies have discussed that the ocean dynamics play an active role in generating AMV (Oelsmann et al., 2020) and its interplay with the NAO (Vecchi et al., 2017; R. Zhang et al., 2019; Kim et al., 2020), although natural and anthropogenic external forcing might be crucial in modulating its amplitude and timing (Bellucci et al., 2017; Bellomo et al., 2018; Andrews et al., 2020; Borchertet al., 2021; Mann et al., 2021; see Sections 3.7.7 and AIV.2.7).

The AR5 assessed with high confidence that initialized predictions can improve the skill for temperature over the North Atlantic, in particular in the sub-polar branch of AMV, compared to the projections, for the first five years (see AR5 WGI Figures 11.3 and 11.4). However, non-initialized predictions showed positive correlation over the same time-range as well, consistent with the notion that part of this variability is caused by external forcing (Section 3.7.7).

Since AR5, near-term initialized predictions, both multi-model (Bellucci et al., 2015a; García-Serrano et al., 2015; D.M. Smith et al., 2019) and single-model ensembles (Marotzke et al., 2016; Simpson et al., 2018; Yeager et al., 2018; Hermanson et al., 2020; Bilbao et al., 2021), confirm substantial skill in hindcasting North Atlantic SST anomalies on a time range of eight to ten years. On the same time range, Borchert et al. (2021) show a substantial improvement in the prediction of the subpolar gyre SST (the northern component of the AMV) in CMIP6 models compared to CMIP5, in both initialized and non-initialized simulations. The higher skill of CMIP6 models can be attributed to a more accurate response of SST variations in the subpolar gyre to natural forcing, possibly originating from the AMOC-related delayed response to volcanic eruptions (Hermanson et al., 2020).

Initialization contributes to the reduction of uncertainty and to predicting subpolar SST amplitude (Borchert et al., 2021). Yet, skill in predicting the AMV is not always translated into equally successful predictions of temperature and precipitation over the nearby land and ocean regions (Langehaug et al., 2017). This might be related to systematic model errors in the simulation of the spatial and temporal structure of the AMV and too weak associated teleconnections (Section 3.7.7), and also to the larger noise in regional land variables compared to the AMV index. However, AMV predictions can be used as proxies to predict other variables such as precipitation over Western Europe and Eurasia and SAT over Mediterranean, Northern Europe and north-east Asia (Årthun et al., 2018; Borchert et al., 2019; Ruggieri et al., 2021) whose relationship with North Atlantic SST is robust in observations, but not well captured in climate models.

Encouraging results about the prediction of land precipitation linked to the warm AMV phase (Section 3.7.7 and Annex IV, Figure AIV.2.7) on a two-to-nine-year time scale are reported in the multi-model study by D.M. Smith et al. (2019). Positive correlations with observations are found in the Sahel, South America, the Maritime Continent. Analyses from large-ensemble decadal prediction systems such as the community Earth system model decadal prediction large ensemble (CESM-DPLE; Yeager et al., 2018) show an improvement with respect to the CMIP5 decadal hindcasts (Martin and Thorncroft, 2014b) in forecasting Sahel precipitation over three to seven years, which is consistent with the current understanding of AMV impact over Africa (Mohino et al., 2016; D.M. Smith et al., 2019). CESM-DPLE predicts drought conditions over the Sahel through 2020, which is compatible with a shift towards a negative phase of AMV as a result of a weakening of the AMOC, advocated by a number of studies (Hermanson et al., 2014; Robson et al., 2014; Yeager et al., 2015).

In summary, the confidence in the predictions of AMV and its effects is medium. However, there is high confidence that the AMV skill over five-to-eight-year lead times is improved by using initialized predictions, compared to non-initialized simulations.

Patterns of near-surface air temperature changes show widespread warming by 2041–2060 and 2081–2100 (Figure 4.19) for all SSPs relative to 1995–2014. The area fraction experiencing warming increases with the level of global mean warming. As GSAT continues to increase, it is very likely that by the middle and the end of the 21st century most of the global land and ocean areas will be warmer than in 1995–2014 (high confidence, Section 4.3.1.1).

Figure 4.19 | Mid-and long-term change of annual mean surface temperature. Displayed are projected spatial patterns of multi-model mean change in annual mean near-surface air temperature (°C) in 2041–2060 and 2081–2100 relative to 1995–2014 for (top) SSP1-2.6 and (bottom) SSP3-7.0. The number of models used is indicated in the top right of the maps. No overlay indicates regions where the change is robust and likely emerges from internal variability, that is, where at least 66% of the models show a change greater than the internal-variability threshold (see Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

The multi-model mean temperature change pattern (Figure 4.19) shows some robust key characteristics that are independent of the time horizon and scenario, such as a land–ocean warming contrast, amplified warming over the Arctic region, assessed below, or the comparatively small warming or even cooling in the North Atlantic subpolar gyre (Section 9.2.1.1). Furthermore, changes in aerosol concentrations and land use and land management can have a direct imprint on the regional warming pattern (Bright et al., 2017; Kasoar et al., 2018). Note that the global average of the pattern shown in Figure 4.19 corresponds to CMIP6 multi-model mean GSAT warming (Section 4.3.1) and is thus somewhat warmer than the warming pattern consistent with the central estimate of the GSAT range assessed in Section 4.3.4. Since the regional mean warming scales well with global warming levels independent of the emissions scenario (Section 4.2.4), the key characteristics of the spatial pattern assessed here are largely independent of the difference between CMIP6 multi-model global mean and assessed global GSAT change.

4.5.1.1.2 Polar amplification

It is very likely that under all SSPs the warming in the Arctic will be more pronounced than in the global average over the 21st century. Since AR5 the understanding of the physical mechanisms driving polar amplification has improved.

The Arctic surface is projected to warm more than the global average over the 21st century, with annual-average Arctic warming of about 3°C (SSP1-2.6), 10°C (SSP3-7.0) and 12°C in (SSP5-8.5) by 2081–2100 relative to 1995–2014 (Figure 4.19). This phenomenon, known as polar or Arctic amplification, is a ubiquitous feature of the response to GHG forcing simulated by climate models (Manabe and Wetherald, 1975, 1980; Manabe and Stouffer, 1980; Robock, 1983; Hansen et al., 1984; Manabe et al., 1991; Holland and Bitz, 2003; Winton, 2006; Pithan and Mauritsen, 2014) and has been observed over recent decades concurrent with Arctic sea ice loss (Section 2.3.2.1; Serreze and Barry, 2011). Based on robust scientific understanding and agreement across multiple lines of evidence (Section 7.4.4.1), there is high confidence that the rate of Arctic surface warming will continue to exceed the global average over the 21st century.

A variety of mechanisms contribute to Arctic amplification (Section 7.4.4.1.1). While surface-albedo feedbacks associated with the loss of sea ice and snow have long been known to play important roles (Arrhenius, 1896; Manabe and Stouffer, 1980; Robock, 1983; Hall, 2004), it is now recognized that temperature (lapse-rate and Planck) feedbacks also contribute to Arctic amplification through a less efficient longwave radiative damping to space with warming at high latitudes (Winton, 2006; Pithan and Mauritsen, 2014; Goosse et al., 2018; Stuecker et al., 2018). Increases in poleward atmospheric latent heat transport and oceanic heat transport also contribute to Arctic warming (Holland and Bitz, 2003; Bitz et al., 2006; Lee et al., 2011, Lee et al., 2017; Alexeev and Jackson, 2013; Marshall et al., 2014, 2015; Woods and Caballero, 2016; Nummelin et al., 2017; Singh et al., 2017; Merlis and Henry, 2018; Oldenburg et al., 2018; Armour et al., 2019; Beer et al., 2020). Projected reduction in the strength of the AMOC over the 21st century is expected to reduce Arctic warming, but even a strong AMOC reduction would not eliminate Arctic amplification entirely (medium confidence) (Liu et al., 2017; Liu et al., 2018; Wen et al., 2018).

There remains substantial uncertainty in the magnitude of projected Arctic amplification (Smith et al., 2020), with the Arctic warming ranging from two to four times the global average in models (Holland and Bitz, 2003; Nummelin et al., 2017). This uncertainty primarily stems from different representations of polar surface-albedo, lapse-rate, and cloud feedbacks, and from different projected poleward energy transport changes (Holland and Bitz, 2003; Crook et al., 2011; Mahlstein and Knutti, 2011; Pithan and Mauritsen, 2014; Bonan et al., 2018). The magnitude of Arctic amplification may also depend on the mix of radiative forcing agents (Najafi et al., 2015; Sand et al., 2016; Stjern et al., 2019) such as the contribution of ozone depleting substances (Polvani et al., 2020). Tropospheric aerosol emissions tend to reduce simulated Arctic warming over the middle of the 20th century (Gagné et al., 2017b) and consequently aerosol emission reductions in observations and SSP scenarios enhance simulated Arctic warming over recent and future decades (Section 6.4.3; Gagné et al., 2015; Acosta Navarro et al., 2016; Wobus et al., 2016; Wang et al., 2018).

Climate models project a weaker polar amplification in the SH than in the NH under transient warming (Figure 4.19). Model simulations (Hall, 2004; Danabasoglu and Gent, 2009; Li et al., 2013) and paleoclimate proxies indicate polar amplification in both hemispheres near equilibrium, but generally with less warming in the Antarctic than the Arctic (Section 7.4.4.1.2). The primary driver of delayed warming of the southern high latitudes is the upwelling in the Southern Ocean and associated ocean heat uptake that is then transported away from Antarctica by northward flowing surface waters (Frölicher et al., 2015; Marshall et al., 2015; Armour et al., 2016; Liu et al., 2018), although asymmetries in feedbacks between the poles also play a role (Section 7.4.4.1.1). Changes in westerly surface winds over the Southern Ocean have the potential to affect the rate of sea-surface warming, but there is currentlylow confidence in even the sign of the effect based on a diverse range of climate model responses to wind changes (Marshall et al., 2014; Ferreira et al., 2015; Kostov et al., 2017; Seviour et al., 2019). A substantial increase in freshwater input to the ocean from the Antarctic ice sheet could further slow the emergence of SH polar amplification by cooling the Southern Ocean surface (Bronselaer et al., 2018; Golledge et al., 2019; Schloesser et al., 2019), but this process is not represented in current climate models which lack dynamic ice sheets. Thus, while there is high confidence that the SH high latitudes will warm by more than the tropics on centennial time scales, there is low confidence that such a feature will emerge this century (Section 7.4.4.1).

Seasonal warming patterns

The warming pattern shows distinct seasonal characteristics. The majority of models show a stronger hemispheric winter than summer warming over land poleward of about 55°N and 55°S (Figure 4.20) and thereby a reduced amplitude of the temperature cycle (Dwyer et al., 2012; Donohoe and Battisti, 2013). On the other hand, over most of the subtropics and mid-latitudinal land regions except for parts of Asia, models project stronger warming in hemispheric summer than winter (Donohoe and Battisti, 2013; Santer et al., 2018), leading to an amplification of the seasonal cycle. This phenomenon has been studied particularly in the case of the amplified summer warming over the Mediterranean region (Seager et al., 2014a; Kröner et al., 2017; Brogli et al., 2019).

Figure 4.20 | Difference of surface temperature change between June–July–August (JJA) and December–January–February (DJF). Displayed are spatial patterns of multi-model mean difference in projected warming in JJA minus warming in DJF in 2081–2100 relative to 1995–2014 for (left) SSP1-2.6 and (right) SSP3-7.0. Diagonal lines mark areas where fewer than 80% of the models agree on the sign of change, and no overlay where at least 80% of the models agree. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

4.5.1.1.3 Changes in temperature variability

It has long been recognized that along with mean temperatures also variance and skewness of the temperature distribution may be changing (Gregory and Mitchell, 1995; Mearns et al., 1997). By amplifying or dampening changes in the tail of temperature distribution such changes are potentially highly relevant to extremes (Section 11.3.1) and pose a serious challenge to adaptation measures. Changes in temperature variability can occur from diurnal to multi-decadal time scales and from the local to the global scale with potentially even opposing signals in different seasons and at the different spatial scales

Changes in GSAT variability are poorly understood. Based on model experiments it has been suggested that unforced variability of GSAT tends to decrease in a warmer world as a result of reduced albedo variability in high latitudes resulting from melting snow and sea ice (Huntingford et al., 2013; Brown et al., 2017), but confidence remains low and an observed change has not been detected. An assessment of changes in global temperature variability is inherently challenging due to the interplay of unforced internal variability and forced changes.

Changes in tropical temperature variability may arise from changes in the amplitude of ENSO (Section 4.5.3.2). Over the extratropics, several studies have identified robust large-scale patterns of changes in variability of annual and particularly seasonal mean temperature, including (i) a reduction in mid- to high-latitude winter temperature variability and (ii) an increase in summer temperature variability over land in the tropics and subtropics (Huntingford et al., 2013; Holmes et al., 2016; see Figure 4.21). The multi-ensemble average across seven single-model initial-condition large ensembles projects a consistent reduction in year-to-year December–January–February (DJF) variability around about 50°N–70°N and June–July–August (JJA) variability around 55°S–70°S along the edge of the sea ice- and snow-covered region (Figure 4.21). There is growing evidence that year-to-year and day-to-day temperature variability decreases in winter over northern mid- to high-latitudes (Fischer et al., 2011; De Vries et al., 2012; Screen, 2014; Schneider et al., 2015; Holmes et al., 2016; Borodina et al., 2017; Tamarin-Brodsky et al., 2020) which implies that the lowest temperatures rise more than the respective climatological mean temperatures (medium confidence). Over the NH, reduced high-latitude temperature variability is associated with disproportionally large warming in source region of cold-air advection due to Arctic amplification and land–sea contrast (De Vries et al., 2012; Screen, 2014; Holmes et al., 2016). It has further been argued that a reduction in snow and sea ice coverage from partly to completely snow- and ice-free ocean and land surface would substantially reduce cold-season temperature variability (Gregory and Mitchell, 1995; Fischer et al., 2011; Borodina et al., 2017) and lead to a shortening of the cold season and earlier onset of the warm season (Cassou and Cattiaux, 2016). Mid-latitudinal winter temperature variability is further affected by a complex interplay of a multitude of processes including potential changes in atmospheric circulation, but there is low confidence in the dominant contribution of Arctic warming compared to other drivers (Cross-Chapter Box 10.1).

Figure 4.21 | Percentage change in interannual variability of (left) December–January–February (DJF) and (right) June–July–August (JJA) mean temperature averaged across seven large initial condition ensembles. Average changes across seven single-model, initial-condition large ensembles are shown for RCP8.5 in 2081–2100 (and where not available for 2080–2099) relative to 1995–2014. Standard deviations are calculated across all members of the large ensembles for every given year to avoid inflation due to the underlying trend and then averaged across the period. Changes are averaged across the ensembles MPI-GE (100 members, Maher et al., 2019), CanESM2 (50 members, Kirchmeier-Young et al., 2017), NCAR-CESM (30 members, Kay et al., 2015), GFDL-CM3 (20 members, Sun et al., 2018), GFDL-ESM2M (30 members, Rodgers et al., 2015), CSIRO-Mk3-6-0 (30 members, Jeffrey et al., 2013) and EC-EARTH (16 members, Hazeleger et al., 2010). Also see Deser et al. (2020) for further information on those ensembles. Diagonal lines indicate areas with low model agreement where fewer than 80% of the models agree on the sign of the change, and no overlay areas with high model agreement where at least 80% of the models agree on the sign of the change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

In JJA, the multi-model average projects an increase in year-to-year JJA variability over Central Europe and North America (Figure 4.21). In particular an increase in daily to interannual summer temperature variability has been projected over central Europe as a result of larger year-to-year variability in soil moisture conditions varying between a wet and dry regime and leading to enhanced land–atmosphere interaction (Seneviratne et al., 2006; Fischer et al., 2012; Holmes et al., 2016). Furthermore, the amplified warming in the source regions of warm-air advection due to land–ocean warming contrast and amplified Mediterranean warming (Seager et al., 2014a; Brogli et al., 2019), may lead to disproportionally strong warming of the hottest days and summers and thereby increased variability. Enhanced temperature variability is further projected over some land regions in the subtropics and tropics (Bathiany et al., 2018).

In summary, there is medium confidence that continued warming will regionally lead to increased and decreased year-to-year temperature variability in the extratropics and there is medium confidence that year-to-year temperature variability will decrease over parts of the mid- to high- latitudes of the winter hemisphere.

Section 12.4.3.2 of AR5 assessed that there is high confidence in the overall pattern of projected end of 21st century tropospheric temperature change and that it is very likely that some of the largest warming will occur in the northern high latitudes. They further assessed that proportionately larger warming is likely to occur in the tropical upper troposphere than at the tropical surface, but with medium confidence owing to the relatively large observational uncertainties and contradictory analyses regarding model accuracy in simulating tropical upper tropospheric temperature trends.

CMIP6 projections show warming throughout the troposphere by the end of this century and a mix of warming and cooling in the stratosphere depending on the emissions scenario (Figure 4.22). The patterns of tropospheric temperature change are highly consistent with those derived from earlier generations of climate models as assessed in AR5, AR4 and TAR. In SSP1-2.6, the multi-model mean warming remains below 3°C everywhere in the troposphere except near the surface in the Arctic; this is similar to the findings in AR5 based on CMIP5 models for RCP2.6. In SSP3-7.0, the zonal mean tropospheric warming is also largest in the tropical upper troposphere, reaching more than 5°C, and near the surface in the Arctic where warming exceeds 8°C (Figure 4.22). It is likely that the warmer projected GSAT in the unconstrained CMIP6 model ensemble contributes to larger warming in the tropical upper troposphere and in the Arctic lower troposphere. This assessment is based on the understanding of polar amplification assessed in Chapter 7 (Section 7.4.4.1), and at low latitudes is based on the understanding of moist convective processes as well as the relationship between CMIP5- and CMIP6-simulated surface temperatures and tropical upper tropospheric warming over the historical period (Section 3.3.1.2).

Figure 4.22 | Long-term change of annual and zonal mean atmospheric temperature. Displayed are multi-model mean change in annual and zonal mean atmospheric temperature (°C) in 2081–2100 relative to 1995–2014 for (left) SSP1-2.6 and (right) SSP3-7.0. The number of models used is indicated in the top right of the maps. Diagonal lines indicate regions where less than 80% of the models agree on the sign of the change and no overlay where 80% or more of the models agree on the sign of the change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Projected stratospheric temperature trends are determined by a balance between the major radiative drivers from ozone recovery, rising CO₂ and other greenhouse gases (including stratospheric water vapour) (Maycock, 2016), as well as future changes in the Brewer –Dobson circulation, which can alter the latitudinal pattern of stratospheric temperature trends (Fu et al., 2015, 2019). In the lower stratosphere, the CMIP6 models project a weak cooling in the inner tropics in SSP1-2.6 and a warming at other latitudes (Figure 4.22). There is enhanced lower stratospheric warming over the Antarctic pole owing to the effects of ozone hole recovery on polar temperatures (Maycock, 2016; Solomon et al., 2017). The projected strengthening of the Brewer –Dobson circulation in the future (Hardiman et al., 2014) also affects stratospheric temperature trends, with adiabatic cooling at low latitudes and warming in middle and high latitudes (Fu et al., 2015, 2019). In SSP3-7.0, there is widespread cooling across much of the stratosphere, as expected from the higher GHG emissions, with a smaller warming in the Antarctic lower stratosphere. Owing to the importance of ozone recovery for the radiative balance of the stratosphere, future global and local stratospheric temperature trends do not scale with projected GSAT change.

In summary, new results since AR5 do not generally alter the understanding of projected zonal mean atmospheric temperature changes. There is high confidence in the overall pattern of projected tropospheric temperature changes given its robustness across many generations of climate models. It is furthervery likely that projected long-term tropospheric warming will be larger than the global mean in the Arctic lower troposphere. It is likely that tropical upper tropospheric warming will be larger than at the tropical surface, however with an uncertain magnitude owing to the potentially large role of natural internal variability and differences across models in the simulated free tropospheric temperature response to a given forcing scenario (Section 3.3.1.2). It is very likely that global mean stratospheric cooling will be larger by the end of the 21st century in a pathway with higher atmospheric CO₂ concentrations.

The AR5 contrasted future changes in near-surface relative humidity (RH) over land and ocean, concluding with medium confidence that reductions in near-surface RH over many land areas are likely . The decrease in near-surface RH over most land areas is associated with the larger warming rates over land than over the ocean and is termed the last-saturation-temperature constraint, as explained in AR5.

Since AR5, significant effort has been devoted to understanding the mechanisms for the decrease in near-surface land RH under global warming, and the relevance of RH changes for the land–sea warming contrast and the water cycle. For the near-surface RH decrease over land, both the moisture transport from the ocean and land–atmosphere feedback processes contribute. For changes in specific humidity over land, the moisture transport from the ocean is dominant while the role of evapotranspiration is secondary (Byrne and O’Gorman, 2016; Chadwick et al., 2016). Nevertheless, the changes in near-surface land RH are also strongly influenced by evapotranspiration, which is suppressed by the drying of soils and plant responses to increasing CO₂ related to stomatal closure under climate change (Byrne and O’Gorman, 2015; Berg et al., 2016; Chadwick et al., 2016; Swann et al., 2016; Lemordant et al., 2018). The combination of oceanic and continental influences can explain the spatially diverse trends in the near-surface RH over land in the observations for the recent decades, with a generally dominant negative trend at the global scale (Vicente-Serrano et al., 2018). There is a strong feedback between the near-surface land RH decrease and land–ocean warming contrast under future warming projections (Section 4.5.1.1).

Changes in land RH can modulate the response of the water cycle to global warming (Chadwick et al., 2013; Byrne and O’Gorman, 2015). Most CMIP5 models project higher precipitation associated with higher near-surface RH and temperature under climate change (Lambert et al., 2017). Over land, the spatial gradients of fractional changes in near-surface RH contribute to a drying tendency in precipitation minus evapotranspiration with warming, which partly explains why the ‘wet gets wetter, dry gets drier’ paradigm does not hold over land (Byrne and O’Gorman, 2015). Terrestrial aridity is projected to increase over land, as manifested by a decrease in the ratio of precipitation to potential evapotranspiration, in which the decrease in near-surface land RH has a contribution of about 35% in CMIP5 models under doubled CO₂ forcing (Fu and Feng, 2014). The aridity can be further amplified by the feedbacks of projected drier soils on land surface temperature, RH, and precipitation (Berg et al., 2016).

The CMIP6 multi-model ensemble projects general decreases in near-surface relative humidity over a large fraction of land areas, but moderate increases over the ocean (Figure 4.23). The projected changes depend on emissions scenario and season. Changes in near-surface RH under SSP1-2.6 are insignificant compared to natural variability. Under SSP3-7.0, during boreal summer, significant decreases relative to natural variability are projected in continental Europe and the Middle East, North America, South America and South Africa.

Figure 4.23 | Long-term changes in seasonal mean relative humidity. Displayed are projected spatial patterns of multi-model mean change (%) in seasonal (top) December–January–February (DJF) and (bottom) June–July–August (JJA) mean near-surface relative humidity in 2081–2100 relative to 1995–2014, for (left) SSP1-2.6 and (right) SSP3-7.0. The number of models used is indicated in the top right of the maps. No overlay indicates regions where the change is robust and likely emerges from internal variability, that is, where at least 66% of the models show a change greater than the internal variability threshold (Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

In summary, there is medium confidence that continued warming will lead to decreased near-surface relative humidity over a large fraction of land areas, but moderate increases over the ocean. There is high confidence that near-surface relative humidity will decrease over parts of the tropical and subtropical latitudes over land.

The AR5 assessed that changes in mean precipitation in a warmer world will exhibit substantial spatial variation. Also, the contrast of mean precipitation between dry and wet regions and between dry and wet seasons will increase over most of globe as temperatures increase. The general pattern of change indicates that high latitude land masses are likely to experience greater amounts of precipitation due to the increased specific humidity of the warmer troposphere as well as increased transport of water vapour from the tropics by the end of this century under the RCP8.5 scenario. Many mid-latitude and subtropical arid and semi-arid regions will likely experience less precipitation, while many moist mid-latitude regions will likely experience more precipitation by the end of this century under the RCP8.5 scenario.

Since AR5, progress has been achieved in understanding changes in patterns and rates of precipitation with GSAT rise. The projected precipitation changes can be decomposed into a part that is related to atmospheric circulation referred to as dynamical component and a part related to water vapour changes, the thermodynamic component. Based on process understanding and modelling (Fläschner et al., 2016; Samset et al., 2016), global mean precipitation will very likely increase by 1–3% per °C of GSAT warming (Section 8.2.1). The increase in atmospheric water vapour is a robust change under global warming, the sensitivity of global precipitation change to warming is smaller (2% per °C) as compared to water vapour change (7% per °C; Held and Soden, 2006). Global energy balance places a strong constraint on the global mean precipitation (Allen and Ingram, 2002; Pendergrass and Hartmann, 2014; Myhre et al., 2018; Siler et al., 2019). Tropospheric radiative cooling constrains global precipitation (Pendergrass and Hartmann, 2014), leading to a slow SST-dependent response and a forcing-dependent rapid adjustment. Rapid adjustments account for large regional differences in hydrological sensitivity across multiple drivers (Samset et al., 2016; Myhre et al., 2017). The rapid regional precipitation response to increased CO₂ is robust across models, implying that the uncertainty in long-term changes is mainly associated with the response to SST-mediated feedbacks (Richardson et al., 2016). Precipitation response to fast adjustments and slow temperature-driven responses are assessed in detail in Chapter 8 (Section 8.2.1).

The thermodynamic response to global warming is associated with a ‘wet get wetter’ mechanism, with enhanced moisture flux leading to subtropical dry regions getting drier and tropical and mid-latitude wet regions getting wetter (Held and Soden, 2006; Chou et al., 2009). Recent studies suggest that the dry-get-drier argument does not hold, especially over subtropical land regions (Greve et al., 2014; Feng and Zhang, 2015; Greve and Seneviratne, 2015). The discrepancy may be partly arising due to differences in model climatologies and by change in the location of wet and dry regions (Polson and Hegerl, 2017). Over the 21st century, significant rate of precipitation change is associated with a spatial stabilization and intensification of moistening and drying patterns (Chavaillaz et al., 2016). In the tropics, weakening of circulation leads to a ‘wet gets drier, dry gets wetter’ pattern (Chadwick et al., 2013). Climate model agreement for precipitation change in the tropics is lower than for other regions (Knutti and Sedláček, 2013; McSweeney and Jones, 2013). Sources of inter-model uncertainty in regional rainfall projections arise from circulation changes (Kent et al., 2015; Chadwick, 2016) and spatial shifts in convection and convergence, associated with SST pattern change and land–sea thermal contrast change (Kent et al., 2015; Chadwick et al., 2017) with a secondary contribution from the response to direct CO₂ forcing (Chadwick, 2016). Factors governing changes in large-scale precipitation patterns are assessed in detail in Sections 8.2.2 and 10.4.1.

Long-term multi-model mean change in seasonal precipitation (JJA and DJF) from CMIP6 models (Figure 4.24) shows substantial regional differences and seasonal contrast. Changes in seasonal precipitation under SSP1-2.6 are small compared to internal variability. Consistent with the AR5, patterns of precipitation change are very likely to increase in the high latitudes especially during local winter and over tropical oceans under SSP3-7.0 (high confidence). CMIP6 projections show an increase in precipitation over larger parts of the monsoon regions and decreases in many subtropical regions including the Mediterranean, southern Africa and south-west Australia (medium confidence). The large-scale patterns of precipitation shown in Figure 4.24 are consistent with the patterns presented in Section 8.4.1.3. Precipitation changes exhibit strong seasonal characteristics (Box 8.2), and, in many regions, the sign of the precipitation changes varies with season. Precipitation variability is projected to increase over a majority of global land area, as assessed in Chapter 8 (Section 8.4.1.3.3), over a wide range of time scales in response to warming (Pendergrass et al., 2017).

Figure 4.24 | Long-term change of seasonal mean precipitation. Displayed are projected spatial patterns of multi-model mean change (%) in (top) December–January–February (DJF) and (bottom) June–July–August (JJA) mean precipitation in 2081–2100 relative to 1995–2014, for (left) SSP1-2.6 and (right) SSP3-7.0. The number of models used is indicated in the top right of the maps. No map overlay indicates regions where the change is robust and likely emerges from internal variability, that is, where at least 66% of the models show a change greater than the internal-variability threshold (Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Most of the projected changes in precipitation exhibit a sharp contrast between land and ocean (Sections 8.2.1 and 8.4.1). Temperature-driven intensification of land-mean precipitation during the 20th century has been masked by fast precipitation responses to anthropogenic sulphate and volcanic forcing (Allen and Ingram, 2002; Richardson et al., 2018a). Based on the Precipitation Driver and Response Model Intercomparison Project (PDRMIP), land-mean precipitation is expected to increase more rapidly with the projected decrease in sulphate forcing and continued warming, contributing to increased global mean precipitation (Table 4.3) and will be clearly observable by the mid-21st century based on RCP4.5 and RCP8.5 scenarios (Richardson et al., 2018a).

Consistent with the findings of AR5, a gradual increase in global mean precipitation is projected over the 21st century with an increase of approximately 2.9% (1.0–5.2%) under SSP1-2.6 and 4.7% (2.3–8.2%) under SSP3-7.0 during 2081–2100 relative to 1995–2014. The corresponding increase in annual mean global land precipitation is 3.3% (0–6.6%), in the SSP1-2.6 and 5.8% (0.5–9.6%) in the SSP3-7.0 (Table 4.3). CMIP6 models show greater increases in precipitation over land than either globally or over the ocean (high confidence).

Based on the assessment of CMIP6 models, we conclude that it is very likely that, in the long term, global mean land and global mean ocean precipitation will increase with increasing GSAT. Annual mean and global mean precipitation will very likely increase by 1–3% per °C GSAT warming. The patterns of precipitation change will exhibit substantial regional differences and seasonal contrast as GSAT increases over the 21st century (high confidence). Precipitation will very likely increase over high latitudes and the tropical ocean and will likely increase in large parts of the monsoon regions. However, it is likely to decrease over the subtropics, including Mediterranean, southern Africa and south-west Australia, in response to GHG-induced warming.

The AR5 assessed changes of the global monsoon in the context of long-term trends across the 21st century and the change by 2081–2100. The AR5 showed growing evidence of improved skill of climate models in reproducing the climatological features of the global monsoon. Taken together with identified model agreement on future changes, the global monsoon precipitation, aggregated over all regional monsoon regions, is likely to strengthen in the 21st century with increases in its area and intensity, while the monsoon circulation weakens. In all RCP scenarios, the global monsoon area is very likely to increase, and the global monsoon precipitation intensity is likely to increase, resulting in avery likely increase in the global monsoon total precipitation, by 2081–2100 (Kitoh et al., 2013).

Since AR5, there has been progress in understanding physical mechanisms for the projected changes in global monsoon and quantifying the sources of uncertainty in projections. The increase in global monsoon precipitation under warming is primarily attributed to the increase of moisture convergence, which comes mainly from the thermodynamic effect due to increasing atmospheric moisture but is partly offset by reduced convergence (W. Zhang et al., 2019; Chen et al., 2020). The dynamic effect, such as monsoon circulation changes, dominates regional differences in the projected monsoon precipitation changes (Chen et al., 2020). Specifically, NH monsoon precipitation will increase more strongly than its SH counterpart, due to an increase in hemispheric temperature difference between the NH and SH, enhancement of the Hadley circulation, and atmospheric moistening, countered by stabilization of the troposphere (Lee and Wang, 2014). The seasonality of global monsoon rainfall is projected to be enhanced in response to warming, featuring a greater wet–dry season contrast (Lee and Wang 2014; Zhang et al. 2019). In addition, the interannual variability of global monsoon rainfall is projected to intensify mainly over land, with a strengthened relationship between global monsoon and ENSO (Hsu et al., 2013; Wang et al., 2020, 2021).

For the uncertainty in mean monsoon precipitation projections, the model uncertainty is the dominant contributor throughout the century and explains more than 70% of the inter-model variance during near term, mid-term, and long term. The contribution of internal variability is only important at the beginning in early decades, while scenario uncertainty becomes important at the end of the 21st century. The sources of uncertainty for the mean and extreme monsoon precipitation mainly differ in the long-term projection, when the contribution of scenario uncertainty is comparable to the model uncertainty for extreme precipitation (Zhou et al., 2020). Although the magnitude of internal variability differs between CMIP5 models and single-model, initial-condition large ensembles, the impact is only evident in the beginning decades. For the mid- and long term, the magnitude difference does not alter that model uncertainty is the dominant source of uncertainty in the projections of global land monsoon precipitation (Zhou et al., 2020).

Based on the projections of changes in precipitation from CMIP6 under the four SSPs, the global monsoon precipitation is likely to strengthen in the 21st century with increases in its intensity, while NH summer monsoon circulation weakens (Figure 4.14). Global land monsoon precipitation will likely increase by 1.3–2.4% per °C GSAT warming among the four scenarios considered here. In the long term, the multi-model mean change (5–95% range of the available 41 projections) of global land monsoon precipitation index is 2.9% (–0.8 to +7.8%), 3.7% (–2.5 to +8.6%), 3.77% (–3.2 to +8.1%), and 5.7% (–2.8 to +12.3%) under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, respectively. This enhancement is caused by thermodynamic responses due to increased moisture, which is partly offset by dynamic responses due to a weakened circulation (Chen et al., 2020). The patterns of monsoon rainfall change in the mid- to long-term include a north–south asymmetry characterized by greater increase in the NH than the SH, and an East–West asymmetry characterized by enhanced Asian-African monsoons and weakened North American monsoon (medium confidence) (Lee and Wang, 2014; Mohtadi et al., 2016; Pascale et al., 2017; Wang et al., 2021).

Based on the assessment of CMIP6 models, we conclude that it is likely that, in the mid- to long term, the global land monsoon precipitation will increase with GSAT rise despite a weakened monsoon circulation. The global land monsoon precipitation will likely increase by 1.3–2.4% per °C GSAT warming among the four scenarios. Monsoon precipitation responses depend on region and emissions scenario (high confidence).

This subsection provides a global overview of long-term changes in atmospheric dynamical features that is complementary to the regional assessment of links to the hydrological cycle in Chapter 8 (Section 8.4.2), and assessment of the connections to extreme events in Chapter 11 (Section 11.7.2).

4.5.1.6.1 Sea level pressure

The AR5 assessed that mean sea level pressure is projected to decrease in high latitudes and to increase in mid-latitudes. Such a pattern is associated with a poleward shift in the storm track and an increase in the annular mode index. This broad pattern is also found in CMIP6 models (Figure 4.25). Under SSP1-2.6, the pattern in sea level pressure change resembles that for SSP3-7.0, but the amplitudes are small compared to internal variability in 20-year means (Figure 4.25). One exception is found in the SH mid-latitudes, where pressure robustly increases in SSP3-7.0 in both austral summer and winter, but shows no robust change in SSP1-2.6. This is likely attributable to the larger GHG forcing in SSP3-7.0 compared to SSP1-2.6, which contributes to a poleward shift of the SH mid-latitude circulation and becomes relatively more important than the effect of ozone recovery which drives an equatorward shift in the circulation (see Section 4.5.3.1 on the Southern Annular Mode; Barnes and Polvani, 2013; Barnes et al., 2014; Bracegirdle et al., 2020b). The poleward shift in SH mid-latitude circulation in SSP3-7.0 likely contributes to the wetting trend at high southern latitudes (Figure 4.25).

Figure 4.25 | Long-term change of seasonal-mean sea level pressure. Displayed are projected spatial patterns of multi-model mean change in (top) December–January–February (DJF) and (bottom) June–July–August (JJA) mean sea level pressure (hPa) in 2081–2100 relative to 1995–2014, for (left) SSP1-2.6 and (right) SSP3-7.0. The number of models used is indicated in the top right of the maps. No overlay indicates regions where the change is robust and likely emerges from internal variability, that is, where at least 66% of the models show a change greater than the internal-variability threshold (Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

As was found in AR5, several regional sea level pressure features stand out from the zonal-mean change. Sea level pressure markedly decreases in north-eastern North America and north-eastern Asia in boreal winter. In boreal summer, sea level pressure robustly decreases in the Mediterranean and the Middle-East, a decrease that has been linked to a large-scale heat low forced by the amplified warming of the region (Haarsma et al., 2009). It is likely that sea level pressure will increase across the south-western North America and Central America in boreal summer under SSP3-7.0 due to an intensification of the eastern North Pacific subtropical summer high (Li et al., 2012) and a weakening of the North American monsoon (Section 4.5.1.5; Pascale et al., 2017; Wang et al., 2020). These changes in circulation are connected to drying across the eastern subtropical Pacific and Central America regions (Figure 4.24).

4.5.1.6.2 Zonal wind and westerly jets

Storm tracks and mid-latitude westerly jets are dynamically related aspects of mid-latitude circulation. The AR5 assessed that a poleward shift of the SH westerlies and storm track is likely by the end of the 21st century under RCP8.5 (medium confidence). In contrast, low confidence was assessed for the storm-track response in the NH.

Under both SSP1-2.6 and SSP3-7.0 there is a strengthening and lifting of the subtropical jets in both hemispheres (Figure 4.26), consistent with the response to large-scale tropospheric warming found in earlier generations of climate models (Collins et al., 2013). In the SH, GHG emissions tend to force a poleward shift of the jet, but this is opposed, particularly in austral summer, by the stratospheric ozone hole recovery (Barnes and Polvani, 2013; Barnes et al., 2014; Bracegirdle et al., 2020b). Consistent with sea level pressure changes, CMIP6 models project a strengthening and poleward shift of the SH jet in austral summer and winter under SSP3-7.0, but smaller and non-robust changes in SH mid-latitude zonal winds under SSP1-2.6 (Figure 4.26; see also Section 4.5.3.1). CMIP6 models show an improved simulation of the SH jet stream latitude (Bracegirdle et al., 2020a; Curtis et al., 2020). This has been linked to a reduction in the projected poleward shift of the SH jet in austral summer compared to the CMIP5 models (Curtis et al., 2020; Goyal et al., 2021), although differences in the pattern of SST response may also play a role (Wood et al., 2020). In the NH extratropics, the changes in lower-tropospheric zonal-mean zonal winds by the end of the century are generally smaller than in the SH. In boreal winter, there is a weak poleward shift of the NH zonal-mean westerly jet maximum in SSP3-7.0.

Figure 4.26 | Long-term change of zonal-mean, zonal wind. Displayed are multi-model mean changes in (left) boreal winter(December–January–February, DJF) and (right) austral winter (June–July–August, JJA) zonal mean, zonal wind (m s^–1) in 2081–2100 for (top) SSP1-2.6 and (bottom) SSP3-7.0 relative to 1995–2014. The 1995–2014 climatology is shown in contours with spacing 10 m s^–1. Diagonal lines indicate regions where less than 80% of the models agree on the sign of the change and no overlay where at least 80% of the models agree on the sign of the change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

CMIP5 and CMIP6 models show a strong seasonal and regional dependence in the response to climate change of NH westerlies (Barnes and Polvani, 2013; Grise and Polvani, 2014b; Simpson et al., 2014; Zappa et al., 2015; Harvey et al., 2020; Oudar et al., 2020). CMIP5 projections indicate a poleward shift of the westerlies in the North Atlantic in boreal summer, while the North Pacific jet weakens in this season (Simpson et al., 2014; Davini and D’Andrea, 2020; Harvey et al., 2020). There is a poleward shift in the westerlies in both the North Pacific and North Atlantic in Autumn (Barnes and Polvani, 2013; Simpson et al., 2014). However, the shift of the westerlies is more uncertain in the other seasons, particularly in the North Atlantic in winter (Simpson et al., 2014; Zappa and Shepherd, 2017). Here, the circulation response is not well described as a simple shift, since the North Atlantic jet tends to be squeezed on both its equatorward and poleward flanks, together with an eastward extension into Europe (Li et al., 2018; Peings et al., 2018; Simpson et al., 2019a; Harvey et al., 2020; Oudar et al., 2020). Simulations indicate that most of the changes in winter storminess over the Euro-Atlantic region will occur only after exceeding the 1.5°C warming level (Barcikowska et al., 2018).

Progress since AR5 has improved understanding of the climate change aspects that can drive these different, and potentially opposite, responses in the mid-latitude jets and storm tracks. A poleward shift of the jets and storm tracks is expected in response to an increase in the atmospheric stratification and in the upper-tropospheric equator-to-pole meridional temperature gradient, while it is opposed by the decrease in the meridional temperature gradient in the lower troposphere associated with the polar amplification of global warming (Harvey et al., 2014; Shaw et al., 2016). Recent analyses have identified additional climate aspects that can drive mid-latitude jet changes, including patterns in sea surface warming (Mizuta et al., 2014; Langenbrunner et al., 2015; Ceppi et al., 2018; Wood et al., 2020), land–sea warming contrast (Shaw and Voigt, 2015), loss of sea ice (Deser et al., 2015; Harvey et al., 2015; Screen et al., 2018b; Zappa et al., 2018), and changes in the strength of the stratospheric polar vortex (Manzini et al., 2014; Grise and Polvani, 2017; Simpson et al., 2018; Ceppi and Shepherd, 2019). From an energetics perspective,the uncertainty in the response of the jet streams depends on the response of clouds, their non-spatially uniform radiative feedbacks shaping the meridional profile of warming (Ceppi et al., 2014; Voigt and Shaw, 2015, 2016; Ceppi and Hartmann, 2016; Ceppi and Shepherd, 2017; Lipat et al., 2018; Albern et al., 2019; Voigt et al., 2019). Climate models seem to underestimate the forced component of the year-to-year variability in the atmospheric circulation, particularly in the North Atlantic sector (Scaife and Smith, 2018), which suggests some relevant dynamical processes may not be well represented. Whether and how this may affect long-term projections is unknown. In conclusion, due to the influence from competing dynamical drivers and the absence of observational evidence, there is medium confidence in a projected poleward shift of the NH zonal-mean low-level westerlies in autumn and summer and low confidence in the other seasons. There is also overall low confidence in projected regional changes in the NH low-level westerlies, particularly for the North Atlantic basin in boreal winter.

The anthropogenic forced signal in extratropical atmospheric circulation may well be small compared to internal variability (Deser et al., 2012b, 2014) and, as assessed in AR5, there is generallylow agreement across models in many aspects of regional atmospheric circulation change particularly in the NH (Shepherd, 2014). The latter means that, in some regions, a multi-model average perspective of atmospheric circulation change represents a small residual after averaging over large intermodel spread. This is in strong contrast to thermodynamic aspects of climate change, such as surface temperature change, for which model results are generally highly consistent (see, e.g., Figure 4.19). Furthermore, models share systematic biases in some aspects of extratropical atmospheric circulation such as mid-latitude jets, which can have complex implications for understanding forced changes (Simpson and Polvani, 2016). Given these issues, an emerging field of research since AR5 has focused on the development of ‘storylines’ for regional atmospheric circulation change (Shepherd, 2019). The storyline approach is grounded in the identification of a set of physical predictors of atmospheric circulation change, such as those described above (Harvey et al., 2014; Manzini et al., 2014; Shepherd et al., 2018), which act together to determine a specific outcome in the projected atmospheric circulation change. The consequences of multi-model spread in the physical predictors of atmospheric circulation change can be investigated, conditioned on a specified level of global warming (see also Section 1.4.4.2 and Box 10.2; Zappa and Shepherd, 2017; Zappa, 2019; Mindlin et al., 2020).

4.5.1.6.3 Storm tracks

As stated in AR5, the number of extratropical cyclones (ETC) composing the storm tracks is projected to weakly decline in future projections, but by no more than a few percent change. The reduction is mostly located on the equatorward flank of the storm tracks, which is associated with the Hadley cell expansion and a poleward shift in the mean genesis latitude of ETCs (Tamarin-Brodsky and Kaspi, 2017). Furthermore, the poleward propagation of individual ETCs is expected to increase with warming (Graff and LaCasce, 2014; Tamarin-Brodsky and Kaspi, 2017), thus contributing to a poleward shift in the mid-latitude transient-eddy kinetic energy. The increased poleward propagation results from the strengthening of the upper tropospheric jet and increased cyclone-associated precipitation (Tamarin-Brodsky and Kaspi, 2017), which are robust aspects of climate change.

In the NH boreal winter, CMIP6 models show a northward shift of the ETC density in the North Pacific, a tripolar pattern in the North Atlantic, and a weakening of the Mediterranean storm track (Figure 4.27a). CMIP6 models show overall low agreement on changes in ETC density in the North Atlantic in boreal winter (Figure 4.27a). A poleward shift of the storm track is evident in the SH (Figure 4.27b), particularly in the Indian and Pacific Ocean sectors. CMIP6 models still feature long-standing biases in the representation of storm tracks; for example, the winter storm track into Europe is too zonal, though different measures of storm track activity indicate some improvements compared to the previous generations of models (Harvey et al., 2020; Priestley et al., 2020).

Figure 4.27 | Changes in extratropical storm track density. Displayed are projected spatial pattern of multi-model mean change of extratropical storm track density in winter (Northern Hemisphere December –January–Februrary, NH DJF, and Southern Hemisphere June–July–August, SH JJA) in 2080–2100 for SSP5-8.5 relative to 1979–2014 based on 13 CMIP6 models. Diagonal lines indicate regions where fewer than 80% of the models agree on the sign of the change and no overlay where at least 80% of the models agree on the sign of change. Units are number density per 5° spherical cap per month. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Regarding the dynamical intensity of the storm tracks (Section 11.7.2), the number of ETCs associated with intense surface wind speeds and undergoing explosive pressure deepening are projected to strongly decrease in the NH winter (Seiler and Zwiers, 2016; Chang, 2018). The weakening of surface winds of ETCs in the NH is attributed to the reduced low-level baroclinicity from SST and sea ice changes (Harvey et al., 2014; Seiler and Zwiers, 2016; J. Wang et al., 2017a). There are, however, regional exceptions such as in the northern North Pacific, where explosive and intense ETCs are projected to increase in association with the poleward shift of the jet and increased upper-level baroclinicity (Seiler and Zwiers, 2016). Eddy kinetic energy and intense cyclone activity are also projected to decrease in the NH summer in association with a weakening of the jet (Lehmann et al., 2014; Chang et al., 2016). However, explosive cyclones tend to be too weak in climate models (Seiler and Zwiers, 2016; Priestley et al., 2020), though this bias seems to be reduced in high-resolution simulations (Jiaxiang et al., 2020). Furthermore, models may not fully capture the contribution of the future increase in mesoscale latent heating to cyclone intensification (Li et al., 2014; Pfahl et al., 2015; Willison et al., 2015; Michaelis et al., 2017). In conclusion, there is only medium confidence in the projected decrease in the frequency of intense NH ETCs.

In contrast to the Northern Hemisphere, the Southern Hemisphere shows an increase in the frequency of intense ETCs in CMIP5 models (Chang, 2017), and there is high confidence that wind speeds associated with ETCs are expected to intensify in the SH storm track for high emissions scenarios. These changes in intensity are accompanied by an overall southward shift of the SH winter storm track (Figure 4.27b) due to the poleward shift in the upper-level jet and the increase in the meridional SST gradient linked to the slower warming of the Southern Ocean (Grieger et al., 2014).

Regardless of dynamical intensity changes, there is high confidence that the number of ETCs associated with extreme precipitation is projected to increase with warming, due to the increased moisture-loading capacity of the atmosphere (Section 8.4.2; Yettella and Kay, 2017; Hawcroft et al., 2018).

4.5.1.6.4 Atmospheric blocking

Blocking is associated with a class of quasi-stationary, high-pressure weather systems in the middle and high latitudes that disrupt the prevailing westerly flow. These events can persist for extended periods, such as a week or longer, and can cause long-lived extreme weather conditions, from heat waves in summer to cold spells in winter (see Section 11.7.2 for a detailed discussion of these features and Section 3.3.3.3 for the assessment of blocking biases in models simulations). The AR5 assessed with medium confidence that the frequency of blocking would not increase under enhanced GHG concentrations, while changes in blocking intensity and persistence remained uncertain.

The CMIP5 projections suggest that the response of blocking frequency to climate change might be quite complex (Dunn-Sigouin et al., 2013; Masato et al., 2013). An eastward shift of winter blocking activity in the NH is indicated (Masato et al., 2013; Kitano and Yamada, 2016; Lee and Ahn, 2017; Matsueda and Endo, 2017) while during boreal summer, blocking frequency tends to decrease in mid-latitudes (Matsueda and Endo, 2017), with the exception of the eastern Europe–western Russia region (Masato et al., 2013). The projected decrease of blocking in boreal summer partially contrasts with the observed increase in Greenland blocking (Hanna et al., 2018; Davini and D’Andrea, 2020). However, as shown in Woollings et al. (2018), the spatial distribution and the magnitude of the suggested changes are sensitive to the blocking detection methods (Schwierz et al., 2004; Barriopedro et al., 2010; Davini et al., 2012). In the SH, blocking frequency is projected to decrease in the Pacific sector during austral spring and summer. However, seasonal and regional changes are not totally consistent across the models (Parsons et al., 2016), and, as assessed in Section 3.3.3.3, model biases might affect their response.

To better understand the uncertainty in future blocking activity, a process-oriented approach has been proposed that aims to link blocking responses to different features of the global warming pattern. Upper-level tropical warming might be the key factor leading to a reduced blocking, because of the strengthening of zonal winds (Kennedy et al., 2016). The more controversial influence of near-surface Arctic warming might lead to an increased blocking frequency (Mori et al., 2014; Francis and Vavrus, 2015) (see Chapter 10, Box 10.1).

Figure 4.28 shows a clear decrease in blocking activity over Greenland and North Pacific for SSP7.0 and SSP8.5. Models with the largest decrease in blocking frequency in boreal winter are those showing the smallest frequency bias during the historical period (Davini and D’Andrea, 2020). In conclusion, there is medium confidence that the frequency of atmospheric blocking events over Greenland and the North Pacific will decrease in boreal winter in the SSP3-7.0 and SSP5-8.5 scenarios.

Figure 4.28 | Projected winter atmospheric blocking frequencies. Box plot showing December –March atmospheric blocking frequencies from historical simulations over 1995–2014 and projections over 2081–2100, over (a) the Central European region (20°W–20°E, 45°N–65°N); (b) the Greenland region (65°W–20°W, 62.5°N–72.5°N); (c) the North Pacific region (130°E–150°W, 60°N–75°N). Values show the percentage of blocked days per season following the (Davini et al., 2012) index. Median values are the thick black horizontal bar. The lower whiskers extend from the first quartile to the smallest value in the ensemble, and the upper whiskers extend from the third quartile to the largest value. The whiskers are limited to an upper bound that is 1.5 times the interquartile range (the distance between the third and first quartiles). Black dots show outliers from the whiskers. The numbers below each bar report the number of models included. Observationally-based values are obtained as the average of the ERA-Interim Reanalysis, the JRA-55 Reanalysis and the NCEP/NCAR Reanalysis. Adapted from Davini and D’Andrea (2020). Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Projections of long-term ocean thermal properties are assessed comprehensively in Chapter 9 (Sections 9.2.1.1 and 9.2.2.1) and are not covered here to avoid unnecessary overlap.

The model-simulated, long-term trend of ocean acidification is assessed in Section 4.3.2.5 and Chapter 5 (Section 5.3.4.1). It is virtually certain that surface ocean acidification will continue in response to the rise in atmospheric CO₂, and continued penetration of anthropogenic CO₂ from the surface to the deep ocean will acidify the ocean interior (Figure 4.29). By the end of this century, under SSP3-7.0, a pH reduction of about 0.3 is found at a few hundred metres depth of the global ocean, with stronger acidification in the interior North Atlantic and the mid- to high-latitude Southern Ocean. At a depth of about 1 km, a pH reduction of about 0.1 is found.

Figure 4.29 | Long-term change of annual and zonal ocean pH. Displayed are multi-model mean change in annual and zonal ocean pH in 2081–2100 relative to themean of 1995–2014 for SSP1-2.6 and SSP3-7.0, respectively. Eleven CMIP6 model results are used. Diagonal lines indicate regions where fewer than 80% of the models agree on the sign of the change and no overlay where at least 80% of the models agree on the sign of change. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Projections with CMIP6 ESMs (Kwiatkowski et al., 2020) show a surface pH decline of –0.16 ± 0.002 (±1 standard deviation) under SSP1-2.6 and –0.44 ± 0.005 under SSP5-8.5 from 1870–1899 to 2080–2099. The high-latitude oceans, in particular the Arctic, show greater decline in pH and accelerated acidification (Terhaar et al., 2020). For the same period, model-projected bottom-water pH decline is –0.018 ± 0.001 under SSP1-2.6 and –0.030 ± 0.002 under SSP5-8.5. The projected large-scale surface ocean acidification will be primarily determined by the pathway of atmospheric CO₂, with weak dependence on change in climate (high confidence) (Section 5.3.4.1; Hurd et al., 2018). However, for a given atmospheric CO₂ scenario, uncertainty in projected ocean acidification increases with ocean depth because of model-simulated differences in ocean circulation that transports anthropogenic CO₂ from the surface to bottom ocean (high confidence) (Kwiatkowski et al., 2020). For example, projected surface pH fully separates between SSPs scenarios before 2050, but some overlap across SSPs is still found for projected bottom-water pH in 2080 (Kwiatkowski et al., 2020).

The AR5 assessed that it is very likely that the El Niño–Southern Oscillation (ENSO) will remain the dominant mode of interannual variability in the future. Moreover, due to increased moisture availability, the associated precipitation variability on regional scales was assessed to likely intensify. An eastward shift in the patterns of temperature and precipitation variations in the North Pacific and North America related to El Niño and La Niña teleconnections was projected with medium confidence. The stability of teleconnections to other regional implications including those in Central and South America, the Caribbean, parts of Africa, most of Asia, Australia and most Pacific Islands were assessed to be uncertain (Christensen et al., 2013).

There is no consensus on changes in amplitude of ENSO SST variability across CMIP iterations. The main factors driving the diversity of ENSO SST amplitude change in climate models are internal variability, SST-mean warming pattern, and model systematic biases. First, pronounced low-frequency modulations of ENSO exist even in unforced control simulations due to internal variability, which leads a large uncertainty in quantifying future ENSO changes (Wittenberg, 2009; Vega-Westhoff and Sriver, 2017; Zheng et al., 2018). Second, ENSO characteristics depend on the climate mean state of the tropical Pacific; however, ENSO can also influence the mean state through non-linear processes (Cai et al., 2015; Timmermann et al., 2018). The response of the tropical Pacific mean state to anthropogenic forcing is characterized by a faster warming on the equator compared to the off-equatorial region, a faster warming of the eastern equatorial Pacific compared to the central tropical Pacific (e.g., El Niño-like mean SST warming, see Section 7.4.4.2), and a weakening of the Walker circulation in most models. Those models with a El Niño-like warming tend to project a strengthening of ENSO SST variability whereas models with a La Niña-like warming tend to project a weakening of variability (Zheng et al., 2016; Kohyama and Hartmann, 2017; J. Wang et al., 2017b; Cai et al., 2018a; Fredriksen et al., 2020). Third, how to take model biases into account leads to different ENSO changes. Kim et al. (2014) suggested that a subset of CMIP5 models that simulate linear ENSO stability realistically exhibit a decrease in ENSO amplitude by the second half of the 21st century. However, an increase of ENSO SST variability has been projected when considering biases in ENSO pattern simulation by different models (Zheng et al., 2016; Cai et al., 2018a). This highlights the importance of constraining tropical Pacific mean state changes in order to enhance confidence in the projected response of ENSO.

There is also no robust consensus on changes in ENSO diversity. Several studies suggest that an increase in Eastern Pacific (EP)-ENSO events tends to be projected particularly in the models with an El Niño-like warming (Zheng et al., 2016; Cai et al., 2018a; Fredriksen et al., 2020). However, Freund et al. (2020) suggested that models with a El Niño-like mean warming show a tendency toward more Central Pacific (CP) events but fewer EP events compared to models with an La Niña-like warming in both CMIP5 and CMIP6 models.

Even though there is low agreement in simulated changes in ENSO SST variability, the majority of models project an increase in amplitude of ENSO rainfall variability attributable to the increase in mean SST and moisture in CMIP5 (Power et al., 2013; Watanabe et al., 2014; Huang and Xie, 2015) and CMIP6 (Yun et al., 2021). It is likely that extreme El Niño events, accompanied by the eastern equatorial Pacific rainfall exceeding the 5 mm day^–1rainfall threshold, will increase in intensity (Cai et al., 2014a, 2017). However, it has also been suggested that historical model biases over the equatorial Pacific cold tongue in CMIP5 may lead to the greater precipitation mean change and amplification of extreme ENSO-associated rainfall in CMIP5 (Stevenson et al., 2021).

There is limited intermodel agreement on future changes in ENSO teleconnections largely depending on changes in the mean state and changes in ENSO properties (Yeh et al., 2018). Many CMIP5 and CMIP6 models project that the centres of the extratropical teleconnection over North Pacific and North America will shift eastward in association with an eastward shift in tropical convective anomalies (Yeh et al., 2018; Fredriksen et al., 2020). There is an indication that tropical cyclones will become more frequent during future El Niño events (and less frequent during future La Niña events) by the end of the 21st century (Chand et al., 2017), thus contributing to the projected increase in ENSO-associated hydro-climate impacts.

While CMIP6 models show no robust change in ENSO SST amplitude in the mid- and long-term period across all four SSPs, a robust increase in ENSO rainfall amplitude is found particularly in SSP2-4.5, SSP3-7.0, and SSP5-8.5 (Figure 4.10). The changes in ENSO rainfall amplitude in the long-term future (2081–2100) relative to the recent past (1995–2014) are statistically significant at the 95% confidence.

To conclude, the forced change in ENSO SST variability is highly uncertain in CMIP5 and CMIP6 models (medium confidence). However, it is very likely that ENSO-related rainfall variability will increase significantly regardless of ENSO amplitude changes in the mid- and long-term future. It is likely that the pattern of ENSO teleconnection over the North Pacific and North America will shift eastward.

In the mid- to long-term, projected climate mean state changes in the tropical Indian Ocean are expected to resemble a positive IOD state, with faster warming in the west compared to the east (Cai et al., 2013; X.-T. Zheng et al., 2013). However, it was argued that this projected mean state change could be due to the large mean state biases in the simulated current climate and potentially not a realistic outcome (G. Li et al., 2016). Mean state biases also lead to lack of consensus on projected equatorial Indian Ocean SST variability and equatorial modes of climate variability independent of the IOD (DiNezio et al., 2020). If mean state change will indeed resemble a positive IOD state, however, this would lead to a reduction in the amplitude difference between positive and negative IOD events, but with no robust change in IOD frequency (Cai et al., 2013). For a small subset of CMIP5 models that simulate IOD events best, a slight increase in IOD frequency was found under the CMIP5 RCP4.5 scenario (Chu et al., 2014).

However, it was also found that the frequency of extreme positive IOD events, which exhibit the largest climate impacts, might increase by a factor of about three under the CMIP5 RCP8.5 scenario (Cai et al., 2014b). Partially consistent with the above result, a more recent study by Cai et al. (2021), based on CMIP5 RCP8.5 and CMIP6 SSP5-8.5 simulations, shows a robust increased SST variability of large positive IOD events, but a decreased variability of moderate IOD events. An approximate doubling of these extreme positive IOD events was still found for global warming of 1.5°C warming above pre-industrial levels, without a projected decline thereafter (Cai et al., 2018b). These results depend, however, on the realism of the projected mean state change in the Indian Ocean (G. Li et al., 2016).

To conclude, the forced change in IOD in mid- and long-term future remains uncertain due to limited lines of evidence and its dependence on model mean biases. However, there is low confidence that the frequency of extreme positive IOD events will increase under the high-emissions scenario of SSP5-8.5.

The AR5 assessed that there is low confidence in projected changes of the Tropical Atlantic Variability (TAV) because of the general failure of climate models to simulate main aspects of this variability such as the northward displaced ITCZ. The models that best represent the Atlantic meridional mode (AMM) show a weakening for future climate conditions. However, model biases in representation of Altantic Niños strongly limit an assessment of future changes.

Long-term changes in TAVs and associated teleconnections are expected as a result of global warming, but large uncertainties exist due to the models’ systematic underestimation of the connection between PDV and Indo-Pacific SST variations (Lübbecke et al., 2018; Cai et al., 2019). Observational analyses show large discrepancies in SST and trade wind strength (Servain et al., 2014; Mohino and Losada, 2015). Single-model sensitivity experiments show that Atlantic Niño characteristics at the end of 21st century remain consistent with those of the 20th century, though changes in the climatological SSTs can lead to changes in the associated teleconnections (Mohino and Losada, 2015).

The weakening of the AMOC expected from global warming (see Section 4.3.2.3) has been suggested to have an influence on the mean background state of tropical-Atlantic surface conditions, thereby enhancing equatorial Atlantic variability and resulting in a stronger tropical Atlantic–ENSO teleconnection (see Section 3.7.5 for a detailed discussion; Svendsen et al., 2014). A recent multi-model study, based on CMIP5, concluded that the TAV-Pacific teleconnection will weaken under global warming due to the increased thermal stability of the atmosphere (F. Jia et al., 2019). However, there is still a clear lack of model studies, and hence no robust evidence on the long-term evolution of TAV and associated teleconnections.

The AR5 assessed that there is low confidence in projections of future changes in Pacific decadal variability (PDV) due to the inability of CMIP5 models to represent the connection between PDV and Indo-Pacific SST variations. Because the PDV appears to encompass the combined effects of different dynamical processes operating at different time scales, representation of PDV in climate models remains a challenge (Section 3.7.6) and its long-term evolution under climate change uncertain.

In addition to uncertainty from the future evolution of the mechanisms that determined the PDV, it is also unclear how the background state in the Pacific Ocean will change due to time-varying radiative forcing, and how this change will interact with variability at interannual and low-frequency time scales (Fedorov et al., 2020). Recent research suggests that the PDV will have a weaker amplitude and higher frequency with global warming (Zhang and Delworth, 2016; Xu and Hu, 2017; Geng et al., 2019). The former appears to be associated with a decrease in SST variability and the meridional gradient over the Kuroshio-Oyashio region, with a reduction in North Pacific wind stress and meandering of the subpolar/subtropical gyre interplay (Zhang and Delworth, 2016). The latter is hypothesized to rely on the enhanced ocean stratification and shallower mixed layers of a warmer climate, which would increase the phase speed of the westward-propagating oceanic waves, hence shortening the decadal to inter-decadal component (Goodman and Marshall, 1999; Zhang and Delworth, 2016; Xu and Hu, 2017). The weakening of the PDV in a warmer climate may reduce the internal variability of global mean surface temperature, to which PDV seems associated (Zhang et al., 1997; Kosaka and Xie, 2016; Geng et al., 2019). Thus, a weaker and higher frequency PDV could reduce the contribution of internal variability to the GSAT trend and eventually lead to a reduced probability of surface-warming hiatus events.

In summary, based on CMIP5, there is medium confidence that a weaker and higher frequency PDV is expected under global warming.

Based on paleoclimate reconstructions and model simulations, AR5 assessed that AMV is unlikely to change its behaviour in the future. However, AMV fluctuations over the coming decades are likely to influence regional climate, enhancing or offsetting some of the effects of global warming.

Recent proxy-derived reconstructions of AMV-related signals show persistent multi-decadal variability over the last three centuries (Kilbourne et al., 2014; Svendsen et al., 2014; Moore et al., 2017), up to the last millennium (Chylek et al., 2011; Zhou et al., 2016; J. Wang et al., 2017b) and beyond (Knudsen et al., 2011). This implies that in the past AMV properties were little affected by large climatic excursions.

AMV long-term changes under future warming scenarios have so far scarcely been investigated. A study on the CMIP5 multi-model simulations under RCP8.5 scenario by (Villamayor et al., 2018) found no substantial differences in the simulated SST patterns (and in the related tropical rainfall response) when RCP8.5, historical and piControl simulations are compared. Such results suggest that the AMV is not expected to change under global warming. A more recent single-model large ensemble study (Hand et al., 2020) shows a pronounced change in the AMV pattern under global warming linked to a strong reduction of the mean AMOC and its variability. However, since a superposition of multiple processes controls the AMV, as extensively discussed in Annex IV (Section AIV.2.7), in Chapter 3 (Section 3.7.7), and in Chapter 9 (Section 9.2.3.1), the length of the RCP8.5 simulations might be not sufficient to properly evaluate the respective weight and interplay of internal components and influences from external forcing on AMV projections.

In conclusion, on the basis of paleoclimate reconstructions and CMIP5 model simulations, there is low confidence that the AMV is not expected to change in the future.

Global warming of 1.5°C implies higher mean temperatures compared to 1850–1900, with generally higher warming over land compared to ocean areas (virtuallycertain) and larger warming in high latitudes compared to low latitudes (Figure 4.31). In addition, global warming of 2°C versus 1.5°C results in robust increases in the mean temperatures in almost all locations, both on land and in the ocean (virtually certain), with subsequent further warming at almost all locations at higher levels of global warming (virtually certain) (Hoegh-Guldberg et al., 2018). For each particular level of global warming, relatively larger mean warming is projected for land regions (virtually certain) (see Figure 4.31; Christensen et al., 2013; Collins et al., 2013; Seneviratne et al., 2016). The projected changes at 1.5°C and 2°C global warming are consistent with observed historical global trends in temperature and their attribution to anthropogenic forcing (Chapter 3), as well as with observed changes under the recent global warming of 0.5°C (Schleussner et al., 2017; Hoegh-Guldberg et al., 2018). That is, spatial patterns of temperature changes associated with the 0.5°C difference in GMST warming between 1991–2010 and 1960–1979 (Schleussner et al., 2017; Hoegh-Guldberg et al., 2018) are consistent with projected changes under 1.5°C and 2°C of global warming.

Figure 4.31 | Projected spatial patterns of change in annual average near-surface temperature (°C) at different levels of global warming. Displayed are (a–d) spatial patterns of change in annual average near-surface temperature at 1.5°C, 2°C, 3°C, and 4°C of global warming relative to the period 1850–1900 and (e–g) spatial patterns of differences in temperature change at 2°C, 3°C, and 4°C of global warming compared to 1.5°C of global warming. The number of models used is indicated in the top right of the maps. No overlay indicates regions where the change is robust and likely emerges from internal variability. That is, where at least 66% of the models show a change greater than the internal-variability threshold (Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. Values were assessed from a 20-year period at a given warming level, based on model simulations under the Tier-1 SSPs of CMIP6. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

The largest increase in annual mean temperature is found in the high latitudes of the Northern Hemisphere across all levels of global warming (virtually certain) (Figure 4.31). This phenomenon peaks in the Arctic and is known as Arctic amplification, with the underlying physical mechanisms assessed in detail in Section 4.5.1 and Chapter 7 (Section 7.4.4.1). For the CMIP6 ensemble average considered here, Arctic annual mean temperatures warm by a factor of 2.3, 2.5, 2.4 and 2.4 for 1.5°C, 2°C, 3°C and 4°C of global warming, respectively. That is, Arctic warming scales approximately linearly with GSAT. Generally, when Arctic amplification is considered across individual models, warming occurs at a factor of two to four times the global level of warming. It is unlikely that warming in the Southern Hemisphere high latitudes in the 21st century will exceed the change in GSAT, or that it will substantially exceed warming in the tropics, for GSAT change ranging between 1.5°C and 4°C (Figure 4.31 and Table 4.2). Correspondingly, there is low confidence of Antarctic amplification occurring under transient, 21st century low mitigation scenarios (Table 4.2 and Section 7.4.4.1). The Antarctic continent is projected to warm at a higher rate than the mid-latitude Southern Ocean, however, at all levels of global warming (Figure 4.31). The relevant physical mechanisms that reduce the amplitude of polar amplification over Antarctica compared to the Arctic are assessed in detail in Section 4.5.1 and Chapter 7 (Section 7.4.4.1). In the Southern Hemisphere the strongest warming over land is to occur, at any given level of global warming, over the subtropical areas of South America, southern Africa and Australia (high confidence). The relatively strong warming in subtropical southern Africa may be attributed to strong soil-moisture–temperature coupling and projected increased dryness under enhanced subsidence (Engelbrecht et al., 2015; Vogel et al., 2017). Across the globe, in the tropics, subtropics, and mid- to high latitudes, temperatures tend to scale linearly with the level of increase in GSAT and patterns of change are largely scenario independent (high confidence).

While global mean precipitation increases as GSAT rises with the very likely range of 1–3% per 1°C (high confidence) (Sections 8.2.1 and 8.4.1), patterns of precipitation change do not scale as linearly with GSAT increase. Nevertheless, common features of precipitation change in the multi-model mean across scenarios still exist for different levels of global warming (Figure 4.32). Precipitation will very likely increase in the high latitudes and over tropical regions, and will likely increase in large parts of the monsoon region, but are likely to decrease over the subtropical regions, including the Mediterranean, southern Africa, parts of Australia and South America at all four levels of global warming. The increases and decreases in precipitation will amplify at higher levels of global warming (high confidence) (Figure 4.32). Changes in extreme precipitation events under different levels of global warming are assessed in Chapter 11.

Figure 4.32 | Projected spatial patterns of change in annual average precipitation (expressed as a percentage change) at different levels of global warming. Displayed are (a–d) spatialpatterns of change in annual precipitation at 1.5°C, 2°C, 3°C, and 4°C of global warming relative to the period 1850–1900. No map overlay indicates regions where the change is robust and likely emerges from internal variability, that is, where at least 66% of the models show a change greater than the internal-variability threshold (Section 4.2.6) and at least 80% of the models agree on the sign of change. Diagonal lines indicate regions with no change or no robust significant change, where fewer than 66% of the models show change greater than the internal-variability threshold. Crossed lines indicate areas of conflicting signals where at least 66% of the models show change greater than the internal-variability threshold but fewer than 80% of all models agree on the sign of change. Values were assessed from a 20-year period at a given warming level, based on model simulations under the Tier-1 SSPs of CMIP6. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

The SR1.5 stated low confidence regarding changes in global monsoons at 1.5°C versus 2°C of global warming, as well as differences in monsoon responses at 1.5°C versus 2°C. Generally, statistically significant changes in regional annual average precipitation are expected at a global mean warming of 2.5°C–3°C or more (Tebaldi et al., 2015). Over the austral-winter rainfall regions of south-western South America, South Africa and Australia, projected decreases in mean annual rainfall showhigh agreement across models and a strong climate change signal even under 1.5°C of global warming, with further amplification of the signal at higher levels of global warming (high confidence) (Mindlin et al., 2020). This is a signal evident in observed rainfall trends over these regions (Sections 2.3.1.3 and 8.3.1.6 ). Also, over the Asian monsoon regions, increases in rainfall will occur at 1.5°C and 2°C of global warming (Chevuturi et al., 2018). At warming levels of 1.5°C and 2°C, the changes in global monsoons are strongly dependent on the modelling strategies used, such as fully coupled transient, fully coupled quasi-equilibrium, and atmosphere-only quasi-equilibrium simulations. In particular, the differences of regional monsoon changes among model setups are dominated by strategy choices such as transient versus quasi-equilibrium set-up, prescription of SST, and treatment of aerosols (Zhang and Zhou, 2021).

The global and land area fractions with significant precipitation changes with global warming are shown in Figure 4.33. It is virtually certain that average warming will be higher over land. As warming increases, a larger global and land area will experience statistically significant increases or decreases in precipitation (medium confidence). The increase of the area fraction with significant precipitation increase is larger over land than over the ocean, but the increase of the area fraction with significant precipitation decrease is larger over the ocean than over land (Figure 4.33). Precipitation variability in most climate models increases over the global land area in response to warming (Pendergrass et al., 2017).

Figure 4.33 | Area fraction of significant precipitation change at 1.5°C, 2°C, 3°C, and 4°C of global warming. Range of land fraction (top) and global area fraction (bottom) with significant precipitation increase (left-hand side) and decrease (right-hand side) in the projected annual precipitation change (%) at levels of global warming compared to the period 1850–1900. Values were assessed from a 20-year period at a given warming level from SSP1-2.6, SSP3-7.0 and SSP5-8.5 in CMIP6. The solid line illustrates the CMIP6-multi model mean and the shaded band is the 5–95% range across models that reach a given level of warming. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

In summary, based on the assessment of CMIP6 models, there is high confidence that global mean precipitation will increase with increase in global mean surface temperature. Precipitation will very likely increase in the high latitudes and over tropical regions, likely increase in large parts of the monsoon region, but will likely decrease over the subtropical regions. There is high confidence that increases and decreases in precipitation will amplify over higher levels of global warming. As warming increases, there is medium confidence that a larger land area will experience statistically significant increases or decreases in precipitation.

The AR5 reported that the application of pattern scaling to extract information on variables other than surface temperature and precipitation was relativelyunexplored. Since AR5, new studies have examined the relationship between projections of mid-latitude atmospheric circulation and GSAT both in terms of interpreting spread in responses across the CMIP5 multi-model ensemble (Grise and Polvani, 2014a, 2016) and to investigate variations in the circulation response as a function of GSAT change over time within a given forcing experiment (Grise and Polvani, 2017; Ceppi et al., 2018).

At a fixed time horizon, the CMIP5 multi-model spread in GSAT explains only a small fraction of the spread in the shift of the Northern Hemisphere mid-latitude circulation due to an abrupt quadrupling in CO₂ (Grise and Polvani, 2016). The fraction of model spread explained by GSAT in the shift of the Southern Hemisphere circulation is larger, but still fairly small (Grise and Polvani, 2014a, 2016). At a fixed time horizon and for a given emissions scenario, CMIP5 multi-model spread in storm track shifts, and the closely related mid-latitude jets, can be better explained by multi-model spread in lower and upper level meridional temperature gradients than by GSAT (Harvey et al., 2014; Grise and Polvani, 2016).

In the North Atlantic, North Pacific, and Southern Hemisphere, the transient response of the mid-latitude jets to forcing behaves non-linearly with GSAT (Grise and Polvani, 2017; Ceppi et al., 2018). This is a consequence of the time-dependence of the relationship between radiative forcing and GSAT and the temporal evolution of SST patterns (Ceppi et al., 2018), with a potential seasonal component in the SH associated with polar stratospheric temperature changes (Grise and Polvani, 2017). Consequently, the epoch approach applied to a transient simulation of the 21st century will overestimate the mid-latitude circulation response in a stabilized climate. Dedicated time slice experiments simulating stabilized climates are therefore required to assess differences in mid-latitude circulation at given levels of global warming (Li et al., 2018). A further complication in the SH is the competing influences of ozone recovery and increasing GHG concentrations on the austral-summer mid-latitude circulation during the first half the 21st century (Barnes and Polvani, 2013; Barnes et al., 2014). Using transient 21st century experiments to diagnose changes in SH mid-latitude circulation at different levels of warming therefore confounds the effects of ozone recovery and GHG increases (Ceppi et al., 2018). Given these various limitations, we do not apply epoch analysis to assess mid-latitude atmospheric circulation changes and related annular modes of variability.

The SR1.5 (IPCC, 2018b) concluded with high confidence that overshoot trajectories ‘result in higher impacts and associated challenges compared to pathways that limit global warming to 1.5°C with no or limited overshoot’. The degree and duration of overshoot affects the risks and impacts likely to be experienced (Hoegh-Guldberg et al., 2018) and the emissions pathway required to achieve it (Akimoto et al., 2018). Consequences relating to ice sheets and climatic extremes have been found to be greater at 2°C of global warming than at 1.5°C (Schleussner et al., 2016; Hoegh-Guldberg et al., 2018) but even on recovery to lower temperatures, these effects may not reverse. Overshoot has been found to lead to irreversible changes in thermosteric sea level (Tokarska and Zickfeld, 2015; Palter et al., 2018; Tachiiri et al., 2019), AMOC (Palter et al., 2018), ice sheets, and permafrost carbon (Sections 4.7.2 and 5.4.9) and to long-lasting effects on ocean heat (Tsutsui et al., 2006). Abrupt changes and tipping points are not well understood, but the higher the warming level and the longer the duration of overshoot, the greater the risk of unexpected changes (Section 4.7.2). Non-reversal of the hydrological cycle has also been found in some studies with an increase in global precipitation following CO₂ decrease being attributed to a build-up of ocean heat (Wu et al., 2010), and to a fast atmospheric adjustment to CO₂ radiative forcing (Cao et al., 2011).

Global temperature is expected to remain approximately constant if emissions of CO₂ were to cease (Section 4.7.1.1), and so reductions in GSAT are only possible in the event of net negative global CO₂ emissions. We assess here results from an overshoot scenario (SSP5-3.4-OS; O’Neill et al., 2016), which explores the implications of a peak and decline in forcing during the 21st century. Reversibility under more extreme and idealized carbon dioxide removal (CDR) scenarios is assessed in Section 4.6.3. In SSP5-3.4-OS, CO₂ peaks at 571 ppm in the year 2062 and reverts to 497 ppm by 2100 – approximately the same level as in 2040. SSP5-3.4-OS has strong net negative emissions of CO₂, exceeding those in SSP1-2.6 and SSP1-1.9 from 2070 onwards and reaching –5.5 PgC yr^–1 (–20 GtCO₂ yr^–1) by 2100. While this causes global mean temperature to decline, changes in climate have not fully reversed by 2100 under this reversal of CO₂ concentration (Figure 4.34). Quantities are compared for 2081–2100 relative to a 20-year period (2034–2053) of the same average CO₂. Differences between these two periods of the same CO₂ are: GSAT: 0.28 ± 0.30°C (mean ± standard deviation); global land precipitation: 0.026 ± 0.011 mm day^–1; September Arctic sea ice area: –0.32 ± 0.53 million km²; thermosteric sea level: 12 ± 0.8 cm. As assessed in Section 9.3.1.1, Arctic sea ice area is linearly reversible with GSAT. Although these climate quantities are not fully reversible, the overshoot scenario results in reduced climate change compared with stabilisation or continued increase in greenhouse gases (Tsutsui et al., 2006; Palter et al., 2018; Tachiiri et al., 2019) (high confidence).

Figure 4.34 | Simulated changes in climate indices for SSP5-3.4-OS plotted against atmospheric CO₂ concentration (ppm) from 480 up to 571 and back to 496 by 2100. (a) Global surface air temperature change; (b) Global land precipitation change; (c) September Arctic sea ice area change; (d) Global thermosteric sea level change. Plotted changes are relative to the 2034–2053 mean which has same CO₂ as 2081–2100 mean (shaded grey bar). Red lines denote changes during the period up to 2062 when CO₂ is rising, blue lines denote changes after 2062 when CO₂ is decreasing again. Thick line is multi-model mean; thin lines and shading show individual models and complete model range. Numbers in square brackets indicate number of models used in each panel. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

The transient climate response to cumulative CO₂ emissions, TCRE, allows climate policy goals to be associated with remaining carbon budgets as global temperature increase is near-linear with cumulative emissions (Section 5.5). Research since AR5 has shown that the concept of near-linearity of climate change to cumulative carbon emissions holds for measures other than just GSAT, such as regional climate (Leduc et al., 2016) or extremes (Harrington et al., 2016; Seneviratne et al., 2016). However, ocean heat and carbon uptake do exhibit path dependence, leading to deviation from the TCRE relationship for levels of overshoot above 300 PgC (Zickfeld et al., 2016; Tokarska et al., 2019). Sea level rise, loss of ice sheets, and permafrost carbon release may not reverse under overshoot and recovery of GSAT and cumulative emissions (Section 4.7). TCRE remains a valuable concept to assess climate policy goals and how to achieve them but given the non-reversibility of different climate metrics with CO₂ and GSAT reductions, it has limitations associated with evaluating the climate response under overshoot scenarios and CO₂ removal (medium confidence).

As CMIP5 and CMIP6 employed different scenario sets (RCPs and SSPs, respectively; see Section 1.6.1.1and Cross-Chapter Box 1.4), we assess how much of the differences in projections are due to the scenario change and how much due to model changes. The CMIP6-simulated GSAT increases tend to be larger than in CMIP5, for nominally comparable scenarios (Section 4.3.1; Tebaldi et al., 2021).

The radiative forcing labels on SSP and RCP scenarios is approximate and enables the multiple climate forcings within the scenario to be characterized by a single number. While the scenarios are similar in terms of the stratospheric adjusted radiative forcing (Tebaldi et al., 2021), they differ more in their effective radiative forcing (ERF). The combination of component forcings (CO₂, non-CO₂ greenhouse gases, aerosols) within the scenario also differ (Meinshausen et al., 2020). The ERF levels in the RCP and SSP scenarios have been calculated by sampling uncertainty in forcing from a range of different GHG species and aerosols (see 7.SM.1.4 for details). Figure 4.35 shows the time evolution and 2081–2100 mean across the families of scenarios and how this affects projections of GSAT. That the ERFs differ between corresponding SSP and RCP scenarios makes a comparison between CMIP6 and CMIP5 projections challenging (Tebaldi et al., 2021). Wyser et al. (2020) find the EC-Earth3-Veg model exhibits stronger radiative forcing and substantially greater warming under SSP5-8.5 than RCP8.5, and similar, but smaller additional warmings for SSP2-4.5and SSP1-2.6 compared with RCP4.5 and RCP2.6, respectively. In addition to the global response, climate can vary regionally due to non-CO₂ components of forcing (Samset et al., 2016; Richardson et al., 2018a, b).

Figure 4.35 | Comparison of RCPs and SSPs run by a single emulator to estimate scenario differences. Time series with 5–95% ranges and medians of (a) effective radiative forcings, calculated as described in Annex 7.A.1; and (b) global surface air temperature projections relative to 1850–1900 for the RCP and SSP scenarios from MAGICC 7.5. Note that the nameplate radiative forcing level refers to stratospheric adjusted radiative forcings in AR5-consistent settings (Tebaldi et al., 2021) while ERFs may differ. MAGICC7.5 is here run in the recommended setup for WGIII, prescribing observed GHG concentrations for the historical period and switching to emissions-driven runs in 2015. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Emulators (Cross-Chapter Box 7.1) can be used to aid understanding of differences between generations of scenarios. The AR5 (Collins et al., 2013) explored the differences between CMIP3 and CMIP5 (their Figure 12.40). Here we use an emulator calibrated to AR6 assessed GSAT ranges, thus eliminating the effect of differences in the model ensembles, to analyse the differences between SSP and RCP scenarios. MAGICC7.5 in its WGIII-calibrated setup (see Cross-Chapter Box 7.1) projects differences in 2081–2100 mean warming between the RCP2.6 and SSP1-2.6 scenarios of around 0.2°C, between RCP4.5 and SSP2-4.5 ofaround 0.3°C and between RCP8.5 and SSP5-8.5 of around 0.3°C (Figure 4.35b). The SSP scenarios also have a wider 5–95% range simulated by MAGICC7.5 explaining about half of the increased range seen when comparing CMIP5 and CMIP6 models. Higher climate sensitivity is, though, the primary reason behind the upper end of the warming for SSP5-8.5 reaching 1.5°C higher than the CMIP5 results. Compared with the differences between the CMIP5 and CMIP6 multi-model ensembles for the same scenario pairs (Table A6 in Tebaldi et al., 2021), the higher ERFs of the SSP scenarios contribute approximately half of the warmer CMIP6 SSP outcomes (medium confidence).

In summary, there is medium confidence that about half of the warming increase in CMIP6 compared to CMIP5 is due to higher climate sensitivity in CMIP6 models; the other half arises from higher ERF in nominally comparable scenarios (e.g., RCP8.5 and SSP5-8.5).

Reducing GHG emissions will eventually slow and limit the degree of climate change relative to high-emissions scenarios such as SSP5-8.5 (very high confidence). Even when CO₂ emissions are reduced, however, atmospheric CO₂ concentrations continue to increase as long as emissions exceed removal by sinks (Millar et al., 2017). Surface warming would likewise initially continue under scenarios of decreasing emissions, resulting in a substantial lag between a peak in CO₂ emissions and peak warming (high confidence) (Ricke and Caldeira, 2014; Zickfeld and Herrington, 2015). The lag between peak emissions and warming depends on the emissions history prior to the peak and also on the rate of the subsequent emissions reductions (Matthews, 2010; Ricke and Caldeira, 2014; Zickfeld and Herrington, 2015).

In addition to the lag between peak emissions and peak warming,the climate response to reduced emissions would be overlain by internal variability, which can amplify or attenuate the forced response. The resulting masking of differences between scenarios is illustrated in Figure 4.36 for GSAT trends over 2021–2040 (Maher et al., 2020). The overall trends conform to expectations in that most simulations show warming almost everywhere, especially under scenario RCP8.5 (Figure 4.36 bottom row). But any individual grid point can in principle show no warming or even cooling, even under RCP 8.5, over the near term (Figure 4.36, middle row). The magnitude of pointwise maximum and minimum temperature trends can be as large as 0.5^°C per year (Figure 4.36, top and middle rows), exceeding possible trends in the global mean by one order of magnitude. While it is only a small fraction of the surface that simultaneously can show cooling, cooling at any given location is fully consistent with globally averaged surface warming over the near term (high confidence, since the findings of Maher et al. (2020) are consistent across six different large initial-condition ensembles).

Figure 4.36 | Masking of climate response to mitigation by internal variability in the near term. Near-term (2021–2040) pointwise maximum (top row) and pointwise minimum (middle row) surface air temperature trends in the large initial-condition ensemble from MPI (left and centre columns), and CESM (right-hand column) models in the RCP2.6 (left-hand column) and RCP8.5 scenarios (centre and right columns). The percentage of ensemble members with a warming trend in the near term is shown in the bottom panels. Figure modified from Maher et al. (2020). Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

An important development since AR5 has been the quantification of when the climate response to mitigation can be expected to emerge from the background noise of internal variability (illustrated in Figure 4.36; see Section 1.4.2.2 and Glossary). A basic ambiguity arises because once mitigation measures are in place, it is no longer possible to observe what the climate would have been without these measures, and any statement about emergence of the response to mitigation is contingent upon the assumed strength of mitigation in relation to an assumed (‘counterfactual’) no-mitigation scenario. Still, there is high agreement on the emergence of the climate response to mitigation across a number of independent studies using different models and different statistical approaches.

Among global quantities, emergence of the response to differing CO₂ emissions – representing differences between low- and high-emissions scenarios – is first expected to arise in global mean CO₂ concentrations, about 10 years after emissions pathways have started diverging (high confidence) (Tebaldi and Friedlingstein, 2013; Peters et al., 2017; Schwartzman and Keeling, 2020; Spring et al., 2020). In these studies, emergence is generally defined as the time at which the global mean concentration first differs between mitigation and non-mitigation scenarios by more than two standard deviations of internal variability, although there are some methodological differences.

Emergence in GSAT would be delayed further, owing to the inertia in the climate system. Although not investigating emergence as defined here in AR6, Tebaldi et al. (2021) used a 20-year running-mean GSAT and compared pairwise either model-by-model or between CM IP6 ensemble means from the core set of five scenarios assessed in this chapter. Differences by more than 0.1°C showed up in most cases in the near term, with only some of the individual models and the comparisons of the closest scenarios showing a delay until the mid-term. Taking internal variability explicitly into account, Tebaldi and Friedlingstein (2013) and Samset et al. (2020) found emergence of mitigation benefits in GSAT changes about 25–30 years after RCP2.6 emissions diverge from the higher-emissions trajectories in RCP4.5 and RCP8.5. Consistently, Marotzke (2019) found about one-third likelihood that a trend reduction in GSAT, over the period 2021–2035 relative to 2005–2020, would be attributable to the emissions reductions implied by the difference between RCP2.6 and RCP4.5. Emergence of the GSAT response to mitigation of individual short-lived climate forcers (SLCFs) would likewise not occur until several decades after emissions trajectories diverge, owing to the relatively small influence of individual SLCFs on the total ERF (Samset et al., 2020), see also Section 4.4.4 and Figure 4.18.

In contrast to the earlier studies, emergence in GSAT within the near-term has recently been found by McKenna et al. (2021) who investigated the likelihood that under the SSP scenarios GSAT trends will exceed the largest historical observed 20-year trends. They found that under scenario SSP1-1.9, the 20-year GSAT trends would likely be lower than in SSP3-7.0 and SSP5-8.5 within the near term. This earlier diagnosed time of emergence compared to Marotzke (2019), while using a similar statistical approach, presumably arose because of the longer-period trends (20 rather than 15 years) and the larger difference between emissions trajectories considered (medium confidence). Using 20-year temperature anomalies relative to 1995–2014 instead of 20-year trends yielded a low probability of emergence (McKenna et al., 2021), consistent with the AR5 (Collins et al., 2013; Kirtman et al., 2013; Tebaldi and Friedlingstein, 2013; Samset et al., 2020). It is not yet understood why GSAT trends appear to show faster emergence of mitigation benefits, compared to GSAT anomalies.

Emergence of mitigation benefits has been studied much less for quantities other than globally and annually averaged CO₂ concentration and surface temperature. Boreal-winter temperatures are more challenging for emergence, due to larger variability in boreal winter and adding a decade to the time of emergence, whereas emergence times for boreal-summer averages are similar to the annual temperature averages (Tebaldi and Friedlingstein, 2013). Emergence happens later at the regional scale, with a median time of emergence of 30–45 years after emissions paths separate in RCP2.6 relative to RCP4.5 and RCP8.5; a stricter requirement of 95% confidence level instead of median induces a delay of several decades, bringing time of emergence toward the end of the 21st century at regional scales (Tebaldi and Friedlingstein, 2013).

Attribution to emissions reductions, for the case of RCP2.6 relative to RCP4.5, is not substantially more likely for 2021–2035 trends in upper-2000 m OHC than for GSAT (Marotzke, 2019), although OHC change is thought to be less susceptible to internal variability. Furthermore, Marotzke (2019) found only around 10% likelihood of mitigation-benefit emergence during 2021–2035 for change in AMOC and September Arctic sea ice area. Tebaldi and Wehner (2018) showed that the differences in temperature extremes between RCP4.5 and RCP8.5 over all land areas become statistically significant by 2050. The seemingly contrasting result of Ciavarella et al. (2017) that mitigation benefits arise earlier for climate extremes poses no contradiction, because Ciavarella et al. (2017) did not look at emergence as defined here but at the extremes of a distribution, which differ between scenarios already at a time when the distributions are still largely overlapping.

In summary, if strong mitigation is applied from 2020 onward as reflected in SSP1-1.9, its effect on 20-year trends in GSAT would likely emerge during the near term, measured against an assumed non-mitigation scenario such as SSP3-7.0 and SSP5-8.5. However, the response of many other climate quantities to mitigation would be largely masked by internal variability during the near term, especially on the regional scale (high confidence). The mitigation benefits for these quantities would emerge only later during the 21st century (high confidence). During the near term, a small fraction of the surface can show cooling under all scenarios assessed here, so near-term cooling at any given location is fully consistent with globally averaged surface warming (high confidence).

CDR options include afforestation, soil carbon sequestration, bioenergy with carbon capture and storage (BECCS), wet land restoration, ocean fertilization, ocean alkalinisation, enhanced terrestrial weathering and direct air capture and storage (see Section 5.6.2 and Table 5.9 for a more complete discussion). Chapter 8 (Section 8.4.3) assesses the implications of CDR for water cycle changes. The potential of different CDR options in terms of the amount of CO₂ removed per year from the atmosphere, costs, co-benefits and side effects of the CDR approaches are assessed in SR1.5 (de Coninck et al., 2018), the AR6 WGIII Report (see Chapters 7 and 12), and in several review papers (Fuss et al., 2018; Lawrence et al., 2018; Nemet et al., 2018). In the literature, CDR options are also referred to as ‘negative CO₂ emissions technologies’.

Deployment of CDR will lead to a reduction in atmospheric CO₂ levels only if uptake by sinks exceeds net CO₂ emissions. Hence, there could be a substantial delay between the initiation of CDR and net CO₂ emissions turning negative (van Vuuren et al., 2016), and the time to reach net negative CO₂ emissions and the evolution of atmospheric CO₂ and climate thereafter would depend on the combined pathways of anthropogenic CO₂ emissions, CDR, and natural sinks. The cooling (or avoided warming) due to CDR would be proportional to the cumulative amount of CO₂ removed from the atmosphere by CDR (Tokarska and Zickfeld, 2015; Zickfeld et al., 2016), as implied by the near-linear relationship between cumulative carbon emissions and GSAT change (Section 5.5).

Emissions pathways that limit globally averaged warming to 1.5°C or 2°C by the year 2100 assume the use of CDR approaches in combination with emissions reductions to follow net negative CO₂ emissions trajectory in the second half of this century. For instance, in SR1.5, all analysed pathways limiting warming to 1.5°C by 2100 with no or limited overshoot include the use of CDR to some extent to offset anthropogenic CO₂ emissions and the median of CO₂ removal across all scenarios was 730 GtCO₂ in the 21st century (Rickels et al., 2018; Rogelj et al., 2018b). Affordable and environmentally and socially acceptable CDR options at scale well before 2050 are an important element of 1.5°C-consistent pathways especially in overshoot scenarios (de Coninck et al., 2018). The required scale of removal by CDR can vary from 1–2 GtCO₂ yr^–1year from 2050 onwards to as much as 20 GtCO₂ yr^–1 (Waisman et al., 2019). In the SSP class of scenarios, net CO₂ emissions turn negative from around 2050 in SSP1-1.9 and around 2070 in SSP1-2.6 and in the overshoot scenario SSP5-3.4-OS (O’Neill et al., 2016). Thus, CDR would play a pivotal role in limiting climate warming to 1.5°C or 2°C (Minx et al., 2018). In stark contrast, however, two extensive reviews (Lawrence et al., 2018; Nemet et al., 2018) conclude that it is implausible that any CDR technique can be implemented at the scale needed by 2050.

When CDR is applied continuously and at scales as large as currently deemed possible, under RCP8.5 as the background scenario, the widely discussed CDR options such as afforestation, ocean iron fertilization and surface ocean alkalinisation are individually expected to be relatively ineffective, with limited (8%) warming reductions relative to the scenario with no CDR option (Keller et al., 2014). Hence, the potential role that CDR will play in lowering the temperature in high-emissions scenarios is limited (medium confidence). The challenges involved in comparing the climatic effects of various CDR options has also been recognized in recent studies (Sonntag et al., 2018; Mengis et al., 2019). For instance, due to compensating processes such as biogeophysical effects of afforestation (warming from albedo decrease when croplands are converted to forests) more carbon is expected to be removed from the atmosphere by afforestation than by ocean alkalinisation to reach the same global mean cooling.

The climate response to CDR-caused net negative CO₂ emissions has been studied in Earth system models by prescribing idealized ramp-down of CO₂ concentrations (MacDougall,2013; Zickfeld et al., 2016; Schwinger and Tjiputra, 2018), CO₂ concentrations of RCP scenarios that have net negative CO₂ emissions (C.D. Jones et al., 2016b), and idealized net negative CO₂ emissions scenarios (Tokarska and Zickfeld, 2015). The Carbon Dioxide Removal Model Intercomparison Project (CDRMIP) uses multiple ESMs to explore the climate response, effectiveness of CO₂ removal, and challenges of CDR options (Keller et al., 2018). Idealized CDRMIP simulations increase CO₂ concentrations at 1% per year from the level in the pre-industrial control run (piControl) to 4×CO₂ and subsequently decrease at the same rate to the piControl level. This section assesses the lag in climate response to CDR-caused negative emissions; climate ‘reversibility’ is assessed in Section 4.7.2. The ramp-down phase, though unrealistic, represents the ‘net negative CO₂ emissions’ phase.

Figure 4.37 illustrates the first results from CDRMIP (Keller et al., 2018). Other studies that use similar (Zickfeld et al.,2016; Schwinger and Tjiputra, 2018; Jeltsch-Thömmes et al., 2020) or other idealized scenarios (MacDougall, 2013) or more realistic net negative CO₂ emissions scenarios such as RCP2.6 (C.D. Jones et al., 2016b) and scenarios that limit warming to 2°C or less after different levels of overshoot (Tokarska and Zickfeld, 2015) arrive at similar conclusions. Changes in key climate variables substantially lag behind the decline in CO₂ (Figure 4.37). The precipitation increase at the beginning of the ramp-down phase agrees with the increase in precipitation for an abrupt decline in CO₂ (Cao et al., 2011). Notwithstanding a decline in atmospheric CO₂, global mean thermosteric sea level would continue to rise. When atmospheric CO₂ returns to the piControl level, global mean thermosteric sea level is higher than its value at peak CO₂ (Figure 4.37), and it is likely that thermosteric global sea level would not return to piControl levels for over 1000 years after atmospheric CO₂ is restored to piControl concentrations (Tokarska and Zickfeld, 2015; Ehlert and Zickfeld, 2018). Therefore, there is high confidence that sea level rise will not be reversed by CDR at least for several centuries (Chapter 9 (Section 9.6.3.5). A comparison of different models shows recovery of AMOC intensity during net negative CO₂ emissions, but the results are model dependent – strengthening with an overshoot in most models (Jackson et al., 2014) and strengthening but not reaching the initial state in some models (Sgubin et al., 2015). The overall lag in response is qualitatively similar to the lagged climate system response in the overshoot scenario SSP5-34-OS where CO₂ rises until 2062 and decreases thereafter (Figure 4.34). The lag in climate response to CDR causes hysteresis between key climate variables such as temperature, precipitation, AMOC and sea level, and atmosphere CO₂ with the hysteresis characteristics dependent on the rate of CDR and climate sensitivity (MacDougall, 2013; Jeltsch-Thömmes et al., 2020).

Figure 4.37 | Delayed climate response to carbon dioxide removal (CDR)-caused net negative CO₂ emissions. Multi-model simulated response in global and annual mean climate variables for a ramp-up followed by ramp-down of CO₂. Atmospheric CO₂ increases from the pre-industrial level at a rate of 1% yr^–1 to 4×CO₂, then decreases at the same rate to the pre-industrial level and then remains constant. The ramp-down phase represents the period of net negative CO₂ emissions. (a) Normalized ensemble mean anomaly of key variables as a function of year, including atmospheric CO₂, surface air temperature, precipitation, thermosteric sea level change (see Glossary), global sea ice area, Northern Hemisphere sea ice area in September, and Atlantic meridional overturning circulation (AMOC); (b) surface air temperature; (c) precipitation; (d) September Arctic sea ice area; (e) AMOC; (f) thermosteric sea level; five-year running means are shown for all variables except the sea level change. In (b, f), red lines represent the phase of CO₂ ramp-up, blue lines represent the phase of CO₂ ramp-down, brown lines represent the period after CO₂ has returned to pre-industrial level, and black lines represent the multi-model mean. For all of the segments in (b, f), the solid coloured lines are CMIP6 models, and the dashed lines are other models (i.e., EMICs, CMIP5-era models). Vertical dashed lines indicate peak CO₂ and when CO₂ again reaches pre-industrial value. The number of CMIP6 and non-CMIP6 models used is indicated in each panel. The time series for the multi-model means (b, f) and the normalized anomalies (a) are terminated when data from all models are not available, in order to avoid the discontinuity in the time series. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure

Termination of CDR refers to a sudden and sustained discontinuation of CDR deployment (see Section 4.6.3.3 for termination effects of SRM). The literature on the termination effects of CDR is limited, mostly considering scenarios where CDR implementation is explicit and does not result in net negative CO₂ emissions (Keller et al., 2014; González et al., 2018). In simulations where CDR is applied on the RCP8.5 scenario at scales as large as currently deemed possible, the increase in global mean warming rates following CDR termination are relatively small in comparison to SRM termination (Keller et al., 2014). The exception is artificial ocean upwelling where surface cooling is mainly caused by bringing cold water from the deep ocean; upon termination this causes larger rates of surface warming (Oschlies et al., 2010). When background emissions are as high as in RCP8.5, termination of a large global-scale application of CDR such as ocean alkalinisation for multiple decades could also result in large regional warming rates (up to 0.15°C per year) that are comparable to those caused by termination of SRM (González et al., 2018). In such cases, large amounts of CO₂ would be removed from the atmosphere before termination, and termination would cause a temporal trajectory of atmospheric CO₂ that is parallel to the high-emissions scenario but from an atmosphere with much lower CO₂ levels. Because CO₂ radiative forcing is a logarithmic function of CO₂ concentration, large regional warming rates are simulated in such terminations. Thus, there is high confidence that the climate effect of CDR termination would depend on the amount CO₂ removed by CDR prior to termination and the rate of background CO₂ emissions at the time of termination. See also Chapter 5, Table 5.9, which summarizes the termination effects of individual CDR options.

In summary, there is high confidence that, due to the near-linear relationship between cumulative carbon emissions and GSAT change, cooling or avoided warming due to a CDR option would depend on the cumulative amount of CO₂ removed by that CDR option. The climate system response to the deployment of CDR is expected to be delayed by years (e.g., in temperature, precipitation, sea ice extent) to centuries (e.g., sea level and AMOC) (high confidence). The climate response to a sudden and sustained CDR termination would depend on the amount of CDR-induced cooling prior to termination and the rate of background CO₂ emissions at the time of termination (high confidence).

The zero emissions commitment (ZEC) is the climate change commitment that would result, in terms of projected GSAT, from setting carbon dioxide (CO₂) emissions to zero. It is determined by both inertia in physical climate system components (ocean, cryosphere, land surface) and carbon cycle inertia (see Annex VII). In its widest sense it refers to emissions of all compounds including greenhouses gases, aerosols and their pre-cursors. A specific sub-category of zero emissions commitment is the zero CO₂ emissions commitment, which refers to the climate system response to a cessation of anthropogenic CO₂ emissions excluding the impact of non-CO₂ forcers. Assessment of remaining carbon budgets requires an assessment of zero CO₂ emissions commitment as well as of the transient climate response to cumulative carbon emissions (TCRE; Section 5.5.2).

There is an offset of continued warming following cessation of emissions by continued CO₂ removal by natural sinks (high confidence) (e.g., Matthews and Caldeira, 2008; Solomon et al., 2009; Joos et al., 2013; Ricke and Caldeira, 2014). Some models continue warming by up to 0.5°C after emissions cease at 2°C of warming (Frölicher et al., 2014; Frölicher and Paynter, 2015; Williams et al., 2017), while others simulate little to no additional warming (Nohara et al., 2015). In SR1.5, the available evidence indicated that past CO₂ emissions do not commit to substantial further warming (Allen et al., 2018). A ZEC close to zero was thus applied for the computation of the remaining carbon budget (Rogelj et al., 2018b). However, the available literature consisted of simulations from a small number of models using a variety of experimental designs, with some simulations showing a complex evolution of temperature following cessation of emissions (e.g., Frölicher et al., 2014; Frölicher and Paynter, 2015; Williams et al., 2017).

Here we draw on new simulations to provide an assessment of ZEC using multiple ESMs (Jones et al., 2019) and EMICs (MacDougall et al., 2020). Figure 4.39 shows results from 20 models that simulate the evolution of CO₂ and the GSAT response following cessation of CO₂ emissions for an experiment where 1000 PgC is emitted during a 1% per year CO₂ increase. All simulations show a strong reduction in atmospheric CO₂ concentration following cessation of CO₂ emissions in agreement with previous studies and basic theory that natural carbon sinks will persist. Therefore, there is very high confidence that atmospheric CO₂ concentrations would decline for decades if CO₂ emissions cease. Temperature evolution in the 100 years following cessation of emissions varies by model and across time scales, with some models showing declining temperature, others having ZEC close to zero, and others showing continued warming following cessation of emissions (Figure 4.39). The GSAT response depends on the balance of carbon sinks and ocean heat uptake (MacDougall et al., 2020). The 20-year average GSAT change 50 years after the cessation of emissions (ZEC₅₀) is summarized in Table 4.8. The mean value of ZEC₅₀ is –0.079°C, with 5–95% range –0.34°C–0.28°C. There is no strong relationship between ZEC₅₀ and modelled climate sensitivity (neither ECS nor TCR; MacDougall et al., 2020). It is therefore likely that the absolute magnitude of ZEC₅₀ is less than 0.3°C, but we assess low confidence in the sign of ZEC on 50-year time scales. This is small compared with natural variability in GSAT.

Figure 4.39 | Zero emissions commitment (ZEC). Changes in (a) atmospheric CO₂ concentration and (b) evolution of global surface air temperature (GSAT) following cessation of CO₂ emissions branched from the 1% per year experiment after emissions of 1000 Pg C (Jones et al., 2019; MacDougall et al., 2020). ZEC is the temperature anomaly relative to the estimated temperature at the year of cessation. ZEC₅₀ is the 20-year mean GSAT change centred on 50 years after the time of cessation (see Table 4.8) – this period is marked with the vertical dotted lines. Multi-model mean is shown as thick black line, individual model simulations are in grey. Further details on data sources and processing are available in the chapter data table (Table 4.SM.1).

Open figure