Constraint on regional land surface air temperature projections in CMIP6 multi-model ensemble

The reliability of the near-land-surface air temperature (LSAT) projections from the state-of-the-art climate-system models that participated in the Coupled Model Intercomparison Project phase six (CMIP6) is debatable, particularly on regional scales. Here we introduce a method of constructing a constrained multi-model-ensemble (CMME), based on rejecting models that fail to reproduce observed LSAT trends. We use the CMME to constrain future LSAT projections under the Shared Socioeconomic Pathways 5-8.5 (SSP5-8.5) and 2-4.5 (SSP2-4.5), representing the high and intermediate scenarios. In comparison with the raw (unconstrained) CMIP6 multi-model ensemble (MME) mean, the impact of the observation-based constraint is less than 0.05(o)C 100 years(-1) at a global scale over the second half of 21(st) century. However, the regional results show a wider range of positive and negative adjustments, from -1.0(o)C 100 years(-1) to 1(o)C 100 years(-1) under the SSP5-8.5 scenario. Although amplitude under SSP2-4.5 is relatively smaller, the CMME adjustment is similar to that under SSP5-8.5, indicating the scenario independency of the CMME impact. The ideal 1pctCO2 experiment suggests that the response of LSAT to carbon dioxide (CO2) forcing on regional scales is responsible for the MME biases in the historical period, implying the high reliability of CMME in the 21(st) century projections. The advantage of CMME is that it goes beyond the idea of model democracy assumed in MME. The unconstrained CMIP6 MME may be overestimating the risks of future warming over North America, but underestimating the risks over Asia.

Automated summary

This study introduces a method of constructing a constrained multi-model ensemble to improve the reliability of near-land-surface air temperature projections. The results show that the constrained ensemble reduces the global-scale temperature change by less than 0.05℃/century compared to the raw ensemble. However, the regional results exhibit a wider range of adjustments, indicating the importance of considering regional impacts. Overall, the constrained ensemble improves the reliability of temperature projections.