We have investigated the ENSO response to warming in four different scenarios with realistic and idealized radiative forcing across 20 models. We found that most of the models show enhanced ENSO SST variability in all warming scenarios considered. Moreover, in a high-forcing scenario SSP5-8.5, each model shows a stronger ensemble mean ENSO. Yet, there are large inter-model differences, which play a greater role than scenario uncertainty. The majority of models do show a generally consistent response across different types of warming experiments indicating that the observed response is indeed a result of radiative forcing, and the spread among models is primarily caused by factors other than natural variability.
Interestingly, for the majority of models, gradual CO2-increase SSP scenarios on average show a stronger ENSO response to warming than the abrupt-4xCO2 scenario. In fact, roughly 40% of the models considered show a weakening or no change of ENSO SST variability in this scenario, which implies that caution should be taken when relying on such experiments for making projections of ENSO future changes.
Despite this uncertainty in ENSO SST response, ENSO rainfall variability in the tropics increases universally across all experiments and models, which has important consequences for adaptation and mitigation as changes to floods and droughts may cause more damage than changes to SST itself and as ENSO remote teleconnections depend of latent heat release. The fact that rainfall variability increases robustly with warming is expected given that a warmer atmosphere can hold more water following the Clausius-Clapeyron relation and eastern Pacific warming can reduce the convection barrier for El Niño events (Yun et al. 2021). However, we find that changes in the background state alone cannot fully account for the modeled change in ENSO rainfall variability as changes in SST variability play a critical role as well. This highlights the need to better understand the response of ENSO SST variability in order to improve predictions for ENSO rainfall response.
Extreme events can change drastically given a small change in ENSO SST amplitude and models with a relatively modest change in ENSO SST variability may show up to a doubling of the number of extreme El Niño events. This close connection between extreme events and ENSO SST amplitude is evident even in a relatively short timeframe of 150 years. The fact that the change in ENSO SST variability is closely connected to an increase in extreme events, which, like increases in ENSO rainfall, has important consequences for adaptation and mitigation as extreme events are often associated with more damage for society than a change in mean conditions (Trenberth 2012). Given the sensitivity of extreme events to small changes in ENSO SST, it is crucial to improve projections of ENSO SST amplitude with warming across models and warming experiments.
In attempt to explain the robust strengthening of ENSO in warming scenarios we compute changes in the Bjerknes Stability Index but find it to be a poor predictor for changes in ENSO in a small subset of models (7), which is similar to findings in other studies using the Bjerknes Index for individual models (Manucharyan and Fedorov 2014; Ferrett and Collins 2019). Like Callahan et al. (2021), we find the increased thermodynamic damping to be the most important term in the Bjerknes Index for 5 out of 7 models, yielding a more stable Bjerknes index those models. However, the fact that all models show a stronger ENSO SST variability in the SSP585 scenario points to other effects counteracting this stability increase. For example, here we show that the thermodynamic damping may be overestimated in the models due to a nonlinear relationship between SST and surface energy fluxes. We show that the coupling between SST and surface heat fluxes decrease in warming experiments above a threshold of 1o C SST anomalies in 4 out of 7 models, which could explain why ENSO SST amplitude can increase despite a stronger thermodynamic damping in a linear sense.
MIROC6 and MIROC-ES2L show a decrease in the Bjerknes Stability Index for the Niño3 region as opposed to other models for which the Bjerknes Index is calculated. These models, together with EC-Earth3 and EC-Earth3-Veg, are outliers among the 20 models analyzed in that they show a drastic increase in ENSO amplitude across experiments. Thus, a decrease in stability for the Niño3 region, driven by increases in feedbacks in the eastern Pacific, may explain why some models have a drastic increase in ENSO amplitude, but it cannot explain the robust increase in ENSO across models.
We suggest here that atmospheric noise, including westerly wind bursts, may play a crucial role in driving changes to ENSO, which is supported by a strong correlation between changes to noise and ENSO SST variability as well as other studies (C. Wengel et al. 2018; Lopez et al. 2022), but questions remain whether this noise is in fact driving a stronger ENSO, as ENSO itself can also generate more noise (Kug et al. 2008; Eisenman, Yu, and Tziperman 2005).
This result highlights a problematic gap in our understanding on what drives changes in ENSO in response to CO2 in GCMs, as we cannot understand changes in ENSO in terms of a simple linear Bjerknes stability framework that links those changes to changes in the tropical mean state. This is exemplified by the fact that mean state changes are larger for the abrupt-4xCO2 scenario, and yet by century-end ENSO amplitude increases more in the SSP585 scenario. In fact, 5 or 6 models in this scenario actually show a weaker ENSO. It is feasible that more comprehensive linear stability analyses, computing the full leading eigen modes of the system (e.g. Sévellec and Fedorov 2013 or Fedorov and Philander 2001) could provide more consistent results, but this is yet to be done for coupled GCMs.
Overall, our results point towards a robust increase in ENSO activity in the SSP5-8.5 and SSP1-2.6 scenarios, yet a common mechanism to explain these changes is lacking. Models that show a drastic increase in ENSO amplitude, such as the MIROC6 and MIROC-ES2L models may be driven by stability changes over the Nino3 region in combination with a small change or no change in thermodynamic damping. On the contrary, for models with a moderate change in ENSO, the linear Bjerknes Index decreases – primarily because of increased thermodynamic damping, leading to a more stable system. This suggests that atmospheric noise and/or nonlinear changes may drive a stronger ENSO for these models. On the other hand, for CESM2, which showed a weaker ENSO with 4xCO2, the Bjerknes Index did show the largest reduction, driven by increased thermodynamic damping and a decrease in the thermocline feedback. Yet, the same model shows an increase in ENSO, albeit small, for SSP5-8.5 and SSP1-2.6 relative to piControl.
ENSO amplitude on average also increases for high CO2 scenarios, such as the 1pctCO2 or abrupt-4xCO2 scenario, although the change is less robust across models. Furthermore, in contrast to Wenger et al. (2021) we did not find evidence that models converge to a weaker ENSO in the abrupt 4xCO2 experiments over longer time-scales. In fact, ENSO remains stronger than the control in several models considered even after 1000 years of computation. On the whole, the abrupt 4xCO2 scenario produces the broadest spread of ENSO projections, and as such may not be the most reliable indicator of changes to come. These findings highlight that despite the robust strengthening of ENSO by century-end in a broad range of models and warming scenarios of CMIP6, there is still a large uncertainty in ENSO future response to global warming, which should be addressed by evaluating ENSO drivers across multiple warming experiments in multiple models.