A multi-model prediction system for ENSO

Abstract

The El Niño and Southern Oscillation (ENSO) is the primary source of predictability for seasonal climate prediction. To improve the ENSO prediction skill, we established a multi-model ensemble (MME) prediction system, which consists of 5 dynamical coupled models with various complexities, parameterizations, resolutions, initializations and ensemble strategies, to account for the uncertainties as sufficiently as possible. Our results demonstrated the superiority of the MME over individual models, with dramatically reduced the root mean square error and improved the anomaly correlation skill, which can compete with, or even exceed the skill of the North American Multi-Model Ensemble. In addition, the MME suffered less from the spring predictability barrier and offered more reliable probabilistic prediction. The real-time MME prediction adequately captured the latest successive La Niña events and the secondary cooling trend six months ahead. Our MME prediction has, since April 2022, forecasted the possible occurrence of a third-year La Niña event. Overall, our MME prediction system offers better skill for both deterministic and probabilistic ENSO prediction than all participating models. These improvements are probably due to the complementary contributions of multiple models to provide additive predictive information, as well as the large ensemble size that covers a more reasonable uncertainty distribution.