Predictability of El Nino Duration Based on the Onset Timing

Analysis of observational data and a long control simulation of the Community Earth System Model, version 1 (CESM1), shows that El Nino events developing in boreal spring to early summer usually terminate after peaking in winter, whereas those developing after summer tend to persist into the second year. To test the predictability of El Nino duration based on the onset timing, perfect model predictions were conducted for three El Nino events developing in April or September in the CESM1 control simulation. For each event, 30-member ensemble simulations are initialized with the same oceanic conditions in the onset month but with slightly different atmospheric conditions and integrated for 2 years. The CESM1. successfully predicts the termination of El Nino after the peak in 95% of the April-initialized simulations and the continuation of El Nino into the second year in 83% of the September-initialized simulations. The predictable component of El Nino duration arises from the initial oceanic conditions that affect the timing and magnitude of negative feedback within the equatorial Pacific, as well as from the Indian and Atlantic Oceans. The ensemble spread of El Nino duration, on the other hand, originates from surface wind variability over the western equatorial Pacific in spring following the peak. The wind variability causes a larger spread in the September-initialized than the April-initialized ensemble simulations due to weaker negative feedback in spring. These results indicate potential predictability of El Nino events beyond the current operational forecasts by 1 year.

Automated summary

The study demonstrates that the duration of El Nino events can be predicted based on their onset timing, with initial ocean conditions playing a crucial role in determining whether the event will terminate or persist. The ensemble spread in duration results from variations in surface wind over the western equatorial Pacific following the peak, which leads to different outcomes in September-initialized simulations compared to April-initialized simulations.