Evaluation of CMIP6 Models in Simulating Network-Based Early Warning Signals of El Niño

Network-based early warning signals of El Niño have been recognized for more than a decade, however, it remains unclear whether current climate models can reproduce these signals. Here, we evaluate simulations from both the pre-industrial control and historical experiments of CMIP6 models. While none of the models exhibited skill in either experiment, performance was generally better in the historical runs, suggesting that the inclusion of external forcing may improve model simulations of the early warning signals. Further analysis indicates that some models such as CESM2, FGOALS-g3, and MRI-ESM2-0 may provide potentially useful early warning information for El Niño events, but their warning signals tended to emerge later than those in reanalysis data. Using a new network-based evaluation metric to assess air-sea interactions in the tropical Pacific, we find that model performance in simulating early warning signals is generally linked to their ability to simulate these interactions. This highlights the importance of improving representations of air-sea coupling in current models. For future investigations into the physical mechanisms underlying the network-based early warning signals, CESM2, FGOALS-g3, and MRI-ESM2-0 are recommended due to their relatively better performance compared to the other models considered in this work, although the causes of their delayed signal emergence require further exploration.