Increased multi-year ENSO predictability under greenhouse gas warming accounted by large ensemble simulations and deep learning

The El Niño/Southern Oscillation (ENSO) is the primary internal climatic driver shaping extreme events worldwide^1,2,3. Its intensity and frequency in response to greenhouse gas (GHG) warming has puzzled scientists for years, despite consensus among models about changes in average conditions^4-16. Recent research has shed light on changes not only in ENSO variability^5,7,8,10,13, but also in the occurrence of extreme^{5,6,11,12,13,14} and multi-year El Niño^4,15, and La Niña^9,11,16 events under GHG warming. Here, we investigate potential changes in ENSO predictability associated with changes in ENSO dynamics in the future by using long-range deep-learning forecasts trained on extensive large ensemble simulations of Earth System Models under historical forcings and the future high GHG emissions scenario. Our results show a remarkable increase in the predictability of ENSO events, ranging from 35% to 65% under the high GHG emissions scenario due to reduced ENSO irregularity, supported by a broad consensus among multi-models. Under GHG warming, an El Nino-like warming flattens the thermocline depth with upper ocean stratification. This flattening of the thermocline depth leads to an increased transition frequency between El Niño and La Niña events, driven by strengthened recharge-discharge oscillation with enhanced thermocline feedback and SST responses to zonal wind stress. As a result, ENSO complexity would reduce with increased regularity and reduced skewness, increasing ENSO predictability. These results imply that the future social and economic impacts of ENSO events may be more manageable, despite an expected increase in the frequency of extreme ENSO events.