Climate Sensitivity in CMIP6: some initial findings

Equilibrium climate sensitivity (ECS) defines the sensitivity of the global climate to a radiative perturbation. ECS has been used for 40 years to estimate the sensitivity of global mean surface temperature to a doubling of atmospheric CO2. The “likely” range (66% probability) of ECS has been 1.5 to 4.5°C over this period.

  • In CMIP6 a significant number of Earth system models exhibit ECS values greater than 4.5°C, with 5 models (so far) suggesting an ECS greater than 5°C (figure 1).
  • While such high ECS values are considered highly unlikely, they cannot be ruled out and therefore remain as; low probability, high impact futures.
  • A new paper, to appear soon, suggests models with ECS greater than ~5°C overestimate post-1990 warming rates compared to observations, suggesting their future warming may also be overestimated. A caveat to this finding is that the post-1990 observed warming includes the highly unusual “warming hiatus” from ~ 1999 to 2012, so the observed record may not fully constrain future warming.
  • At least two CMIP6 models (UKESM1/HadGEM3 and CESM2) have higher ECS than predecessor versions due to an increase in the amount of midto-high latitude super-cooled liquid (SCL) clouds simulated for the present-day. SCL clouds form at temperatures below 0°C but are composed primarily of super-cooled liquid droplets rather than ice crystals. Such clouds are most common in clean regions such as the Southern Ocean.
  • Earlier models (as in CMIP5) systematically overestimated (underestimated) the amount of ice (super-cooled liquid) in these clouds. The two CMIP6 models do a better job in this regard and are closer to observations.
  • In CMIP5 models, these (erroneous) ice clouds melt and become liquid clouds as the climate warms in the future. For an equivalent amount of water, changing from ice to liquid leads to these clouds being more reflective to solar radiation (liquid clouds are composed of a larger number of smaller droplets than ice clouds that are composed of fewer, large crystals).
  • Increasing cloud reflectivity as the climate warms is a (cooling) negative feedback. In CMIP5 models, this negative (mid-latitude) cloud feedback opposed other, predominantly (warming) positive, cloud feedbacks in the tropics and subtropics.
  • The negative mid-latitude cloud feedback is likely spurious in CMIP5 models, as they start from an erroneous present-day, ice-rich cloud state, leading to an unrealistic amount of ice to liquid melt as the climate warms.
  • The improved CMIP6 models do not exhibit this spurious negative cloud feedback. Therefore, the positive (tropical) cloud feedbacks that are still present in these models, are no longer balanced by this negative feedback, resulting in a larger positive net cloud feedback and an increase in climate sensitivity.
  • The improved evaluation of mid-latitude clouds in the CMIP6 models makes us confident the reduced negative cloud feedback is more realistic in these models than their predecessor CMIP5 versions.
Figure 1

Figure 1. Range of Climate Sensitivities across different IPCC AR cycles, contrasted with the multi-model range in CMIP5/AR5 and as of September 2019 in CMIP6. Figure made by Manuel Schlund (DLR) using theESMValTool package.

Figure 2

Figure 2. Global mean surface warming relative to 1850-1900 mean for 10 CMIP6 models (left side), following five SSP pathways shown in the legend and the full CMIP5 ensemble (right side, with time-axis reveresed) following the earlier CMIP5 RCPs. The red plume (red line) plots the CMIP6/CMIP5 multi-model ensemble (ensemble mean) of SSP5-85/RCP85, while the blue plume plots the same for SSP1-26/RCP26. The grey plume and black line shows the historical temperature anomaly for CMIP6 and CMIP5. Figure produced by Erich Fischer ETH Zurich using the ESMValTool package.

  • This does not mean the resulting (higher) ECS is also more realistic. It is possible we have (correctly) removed one erroneous feedback, while other, now unbalanced, feedbacks remain that are incorrect in magnitude.
  • While the high ECS values (greater than ~5°C) seen in some CMIP6 models are very likely greater than the Earth’s actual climate sensitivity, for some of these models we are confident the underpinning feedback processes controlling the ECS response are now more realistic than in predecessor models. This gives us confidence our models are improving at the process level.
  • The high ECS that appears as one erroneous feedback is removed helps focus attention onto the remaining key feedbacks that are no longer balanced by this erroneous feedback. Following this route will ultimately lead to models that have the right future feedback response (measured by global climate sensitivity) for the correct underpinning reasons. Such models will best be able to support reliable future decision-making.

Consequences for future warming: CMIP6 scenarioMIP

  • There are numerous scenarios in the literature exploring future changes in socio-economic development and associated greenhouse gas emissions. It is too costly to sample all of these with Earth system models, a representative subset were therefore selected for CMIP6 scenarioMIP.
  • These SSPs (Shared Socio-economic Pathways), mirror the global radiative forcing levels of the RCPs assessed in IPCC AR5, with an additional scenario (SSP1-1.9) that aims to keep global warming below 1.5°C.
  • In contrast to the RCPs, each SSP now has a defined socio-economic developemnt pathway associated with its climate forcing pathway, allowing future climate change to be assessed in combination with known socoioeconomic futures.

  • As of Sept 2019, 15 models had Tier 1 scenarioMIP data available on the Earth System Grid Federation. These include the aforementioned models with high climate sensitivity.

  • Future temperatures in CMIP6 span a broader range than in CMIP5, with the upper range of the multi-model ensemble exceeding that of CMIP5; compare the red CMIP6 SSP5-85 plume (left side of figure 2), to the red RCP85 CMIP5 plume (right side of figure 2). Similar findings hold for the blue plumes; SSP1-26 (CMIP6) and RCP26 (CMIP5).

  • SSP1-2.6 in CMIP6 appears to have a higher chance of exceeding 2°C than the equivalent RCP26 in CMIP5. We stress not all models have high ECS – some of the new models also have low ECS values. However, the CMIP6 generation of models indicate an increased risk that current climate policies may see an overshoot of the 2°C threshold, strengthening the need for rapid and deep CO2 reductions.

  • As more CMIP6 data becomes available it is possible the ensemble mean warming for SSP5-85 and SSP1-26 will decrease slightly, towards CMIP5 results. The upper tail of the distribution will very likely not change.

Using CMIP6 projections for risk assessment

  • Sutton (2018) argues for framing climate change in the context of risk assessment and risk management. Figure 3 summarizes this approach, showing how different climate sensitivities (of varying degrees of likelihood) give rise to different climate impacts, where impact encompasses potential climate outcomes (e.g. increased flooding, droughts etc). Combining likelihood and impact leads to the potential “risk” to society.

  • Low climate sensitivities may be more likely but, due to the more moderate associated climate impacts, the risk to society may also be moderate. In contrast, high climate sensitivities may have a very low likelihood of occurrence, but they are associated with significantly higher potential impacts. Hence, while these high climate sensitivities are low probability, they are also high impact and thus carry significant risk to society.

  • CMIP6 models with high climate sensitivities represent this category of low probability, high impact, high risk futures. It is therefore crrucial future projections made with these models are used in combination with the more likely, lower impact projections, to develop sound adaptation stratgeies encompassing the full range of possible futures risks faced by society.

Implications for carbon budgets

  • The median value of the transient climate response to cumulative carbon emissions (TCRE), which expresses the global mean surface temperature change per unit of emitted CO2, has not changed significantly from CMIP5 to CMIP6 (see figure 3), with the caveat that only a limited number of CMIP6 models that calculate TCRE are presently available. This implies no major change in our central estimate of allowable carbon emissions to stay below 1.5/2°C warming, for a 50% probability of success.

  • As a result of higher ECS and, therefore also higher TCRE values, some CMIP6 models do suggest a reduced carbon budget to stay below 1.5/2°C warming. Therefore, if a higher probability (e.g. 66% likelihood or greater) is required, allowable carbon emissions may be smaller than suggested in IPCC AR5.
Figure 3

Figure 3. A schematic representation of how Likelihoods of a given outcome (e.g .of Climate sensitivity) combines with the impacts associated with each outcome to resuls in a risk to society (Sutton 2018).

Author: Colin Jones (U. Leeds) Contributions from: Chris Jones (Met Office), Cat Scott (U. Leeds),  Laurent Bopp (Ecole Normale and IPSL) and Maarten van den Berg (PBL)