Ocean Dynamic Persistence Provides Substantial Predictability for El Niño-Southern Oscillation

ENSO predictions are primarily based on coupled ocean-atmosphere dynamical models and statistical models. The prediction skill for coupled models, is contributed by the accuracy of the initial ocean state (typically obtained from data assimilation) and the fidelity of the coupled models in representing ocean-atmosphere interaction after the initialization. Coupled-model initialization shock or drift can limit the contribution of an accurate initial ocean to ENSO prediction skill. However, it is not clear how much ENSO predictability the initial ocean state alone can provide without considering subsequent ocean-atmosphere coupling. Here, we address this question by performing a set of 12-month hindcasts of the 3-D ocean state using a global ocean model. The ocean model is initialized from the initial condition at every month and every year from 1992 to 2017 obtained from an ocean state estimation produced by the Estimating the Circulation and Climate of the Ocean (ECCO). During each 12-month hindcast, the atmospheric forcings are set to climatological seasonal forcings. Therefore, any interannual ocean anomaly in the 12-month hindcasts is solely due to the evolution of the initial ocean state without subsequent ocean-atmosphere coupling. We refer to this as ocean dynamic persistence. This contrasts the statistical persistence (a baseline evaluation metrics for predictions) that damps the initial anomaly statistically as a function of prediction lead time. We show that ocean dynamic persistence has much better skill in predicting the Niño3.4 sea surface temperature anomaly (SSTA), an ENSO index, than the statistical persistence. Therefore, ocean dynamic persistence raises the bar for the baseline metrics of evaluating ENSO prediction skill. Moreover, ocean dynamic persistence hindcasts outperform most dynamical and statistical models up to several months of lead time. Our result highlights the substantial predictability of ENSO provided by ocean dynamic persistence alone. It also underscores the importance of reducing coupled model initialization shock and drift to maximize the positive contribution of ocean dynamic persistence to ENSO prediction skill.