Exploring the initial errors that cause a significant “spring predictability barrier” for El Niño events

Abstract

Within the Zebiak‐Cane model, we identify two types of initial errors that have significant season‐dependent evolutions related to the spring predictability barrier (SPB) for El Niño events. One type includes the sea surface temperature anomaly (SSTA) errors that have a zonal dipolar pattern with positive anomalies in the central equatorial Pacific and negative ones in the eastern equatorial Pacific; the other type consists of the SSTA errors with a spatial structure opposite to that of the former type, the zonal dipolar pattern shows negative anomalies in the central equatorial Pacific and positive anomalies in the eastern equatorial Pacific. The patterns of these two types of errors are nearly opposite of each other. The former causes the El Niño event to be underpredicted, and the latter causes the El Niño event to be overpredicted. For strong El Niño events the former tends to have a larger effect on the predictions than the latter, but for weak El Niño events, it is very difficult to determine which type of initial errors results in worse predictions. It is thought that strong (weak) El Niño events could be affected by strong (weak) nonlinearities. There are also other initial errors; however, they do not yield considerable season‐dependent evolutions nor can a common characteristic be extracted from their patterns. The two types of initial errors suggest two dynamical behaviors of error growth related to the SPB: in one case, the initial errors grow in a manner similar to the El Niño events; in the other case, the initial errors develop with a tendency opposite to the El Niño events. The two types of initial errors may capture the errors that exhibit significant season‐dependent evolutions related to the SPB. In addition, they may provide information regarding the “sensitive area” of ENSO predictions because of their localized regions. Therefore, if these types of initial errors exist in the realistic El Niño–Southern Oscillation (ENSO) predictions and if a data assimilation or a target method can filter them, the ENSO forecast skill may be improved. For ensemble forecast studies, different signs of prediction errors caused by the two types of initial errors could illustrate why the ensemble mean offers a better forecast than a single prediction.