Future global warming estimates have been similar across past assessments, but several climate models of the latest Sixth Coupled Model Intercomparison Project (CMIP6) simulate much stronger warming, apparently inconsistent with past assessments. Here, we show that projected future warming is correlated with the simulated warming trend during recent decades across CMIP5 and CMIP6 models, enabling us to constrain future warming based on consistency with the observed warming. These findings carry important policy-relevant implications: The observationally constrained CMIP6 median warming in high emissions and ambitious mitigation scenarios is over 16 and 14% lower by 2050 compared to the raw CMIP6 median, respectively, and over 14 and 8% lower by 2090, relative to 1995–2014. Observationally constrained CMIP6 warming is consistent with previous assessments based on CMIP5 models, and in an ambitious mitigation scenario, the likely range is consistent with reaching the Paris Agreement target.

Between about 1998 and 2012, a time that coincided with political negotiations for preventing climate change, the surface of Earth seemed hardly to warm. This phenomenon, often termed the 'global warming hiatus', caused doubt in the public mind about how well anthropogenic climate change and natural variability are understood. Here we show that apparently contradictory conclusions stem from different definitions of 'hiatus' and from different datasets. A combination of changes in forcing, uptake of heat by the oceans, natural variability and incomplete observational coverage reconciles models and data. Combined with stronger recent warming trends in newer datasets, we are now more confident than ever that human influence is dominant in long-term warming.

The level of agreement between climate model simulations and observed surface temperature change is a topic of scientific and policy concern. While the Earth system continues to accumulate energy due to anthropogenic and other radiative forcings, estimates of recent surface temperature evolution fall at the lower end of climate model projections. Global mean temperatures from climate model simulations are typically calculated using surface air temperatures, while the corresponding observations are based on a blend of air and sea surface temperatures. This work quantifies a systematic bias in model‐observation comparisons arising from differential warming rates between sea surface temperatures and surface air temperatures over oceans. A further bias arises from the treatment of temperatures in regions where the sea ice boundary has changed. Applying the methodology of the HadCRUT4 record to climate model temperature fields accounts for 38% of the discrepancy in trend between models and observations over the period 1975–2014.

Correcing for these biases and accouning for wider uncertainies in radiaive forcing based on recent evidence, we infer an observaion-based best esimate for TCR of 1.66 °C with a 5-95 % range of 1.0-3.3 °C, consistent with the climate models considered in the IPCC 5 th Assessment Report.TCR for the Climate Model Intercomparison Project, phase 5 (CMIP5) models is deined using simulaions in which atmospheric CO 2 increases at 1 % per year and the muli-model mean is 1.8 °C (1.2-2.4 °C, henceforth bracketed values refer to 5-95 % ranges). 6-8 TCR has also been esimated from Earth's energy budget using: eicacy of ocean heat uptake, 20-22 structural uncertainies in energy-budget calculaions or lower real-world TCR.We focus on potenial biases in temperature series due to geographical incompleteness of the data ('masking') and the combinaion of air and water measurements ('blending') by applying energy-budget TCR calculaions to CMIP5 simulaions and observaions. We calculate energy-budget TCR with the Oto et al. (2013) method, henceforth 'Oto', which uses diferences between an early baseline period and a recent reference period: Where Δ Q is the system heat uptake which, being posiive during warming, means that ECS is larger than TCR. We do not calculate ECS here to avoid uncertainies associated with Δ Q , and to avoid the assumpion of linear climate response which is less accurate over the longer ime periods required for equilibrium. 17 However, as Δ T is in the numerators of Equaions (1) and (3), any Δ T bias afects each calculaion equally in percentage terms.Formally, TCR refers to global near-surface air temperature ('tas' in CMIP5 nomenclature) for Δ T while observaional temperature records have incomplete and varying geographical coverage and combine air temperatures over land and sea ice with near-surface water temperatures over oceans.These diferences introduce biases as warming is not spaially uniform, sea ice coverage changes and Model temperatures are reconstructed in three ways: by using global air temperature ('tas-only'), by blending air temperature over land and sea ice with ocean temperatures over water ('blended') and by blending temperatures and using the historical geographical coverage of observaions in HadCRUT4 ('blended-masked'). We assume that the modelled near-surface water temperature over oceans ('tos' in CMIP5 nomenclature) is equivalent to measured sea surface temperatures. Results are similar between models with diferent ocean layering: for example with 2.5 metre top-layer depth instead of 10 metres, suggesing tos is a robust measure of modelled sea surface temperature (see Supplementary Informaion).The 'tas-only' reconstrucions are used in standard model assessments of TCR, the 'blended' reconstrucions represent the same reconstrucion techniques as HadCRUT4 but with perfect data coverage and the 'blended-masked' reconstrucions represent HadCRUT4. Table 8 shows that the masking bias is largely due to undersampling of rapidly warming polar regions. The blending and masking efects were not acc...

Equilibrium climate sensitivity‐the equilibrium warming per CO2 doubling‐increases with CO2 concentration for 13 of 14 coupled general circulation models for 0.5–8 times the preindustrial concentration. In particular, the abrupt 4 × CO2 equilibrium warming is more than twice the 2 × CO2 warming. We identify three potential causes: nonlogarithmic forcing, feedback CO2 dependence, and feedback temperature dependence. Feedback temperature dependence explains at least half of the sensitivity increase, while feedback CO2 dependence explains a smaller share, and nonlogarithmic forcing decreases sensitivity in as many models as it increases it. Feedback temperature dependence is positive for 10 out of 14 models, primarily due to the longwave clear‐sky feedback, while cloud feedbacks drive particularly large sensitivity increases. Feedback temperature dependence increases the risk of extreme or runaway warming, and is estimated to cause six models to warm at least an additional 3K under 8 × CO2.