Long run emulator calibration increases warming and sea-level rise projections

Owing to their short runtime compared to Earth system models (ESMs), as well as the difficulty for the latest ESMs from the Coupled Model Intercomparison Project Phase 6 (CMIP6) to reproduce historical warming and the so-called ‘hot model problem’, constrained reduced-complexity climate models (‘emulators’) are increasingly used to produce global warming projections from emissions scenarios. Emulators are often calibrated on idealised abrupt CO2 quadrupling experiments from CMIP6, particularly the global surface temperature response over time to an imposed radiative forcing. Such CMIP6 experiments tend to be run for 150 years, which is not sufficient to reveal the full equilibrium response to an imposed climate forcing. Here we show that, when longer experiments are available for emulator calibration, the long-term climate warming projections increase, particularly for 2100, by up to 0.70 (0.42–0.93, 25th to 75th percentile) °C in the median under a high emissions scenario; peak global warming in a high overshoot scenario is higher by 0.24 °C (0.14–0.31 °C). Corresponding long-term thermosteric sea level rise (SLR) is consequently higher, by 0.45 (0.22–0.52, 25th to 75th percentile) m in 2500. This result, consistent across calibrations from 17 ESMs, has implications for climate change mitigation strategies, as it is likely that even more stringent emissions reductions would be required to limit long-term warming and SLR than previously thought.