Quantifying the Predictability of ENSO Complexity Using a Statistically Accurate Multiscale Stochastic Model and Information Theory

Abstract

Abstract An information theory–based framework is developed to assess the predictability and quantify the forecast uncertainty of ENSO complexity, which includes different types of ENSO events with diverse characteristics. With the assistance of a recently developed multiscale stochastic conceptual model that successfully captures both the large-scale dynamics and many crucial statistical properties of the observed ENSO complexity, it is shown that different ENSO events possess distinct predictability limits. Beyond the ensemble mean value, the spread of the ensemble members also contains valuable information about predictability. First, La Niña events are most predictable at long lead times, especially as a subsequent transition after eastern Pacific (EP) El Niño events or during multiyear La Niña phases. Second, EP El Niños tend to be more predictable than the central Pacific (CP) El Niño events up to one year ahead due to a more favorable signal-to-noise ratio, even though their onset remains hard to predict. Third, 4 out of 6 CP El Niño events seem to be predictable up to 24 months ahead, where such strong predictability is often converted to skillful forecast. Fourth, strengthening/weakening the Walker circulation intensity increases/decreases CP predictability at long leads. Fifth, accounting for intraseasonal wind events in the initial condition strongly contributes to EP predictability at lead times of less than one year. Finally, it is shown that a Gaussian approximation of the information gain computation is accurate, making the information theory approach tractable for studying the predictability of more sophisticated models.