Contrasting El Nino-La Nina Predictability and Prediction Skill in 2-Year Reforecasts of the Twentieth Century

Despite the growing demand for long-range ENSO predictions beyond 1 year, quantifying the skill at these lead times remains limited. This is partly due to inadequate long records of seasonal reforecasts that make skill estimates of irregular ENSO events quite challenging. Here, we investigate ENSO predictability and the dependency of prediction skill on the ENSO cycle using 110 years of 24-month-long 10-member ensemble reforecasts from ECMWF's coupled model (SEAS5-20C) initialized on 1 November and 1 May during 1901-2010. Results show that Nino-3.4 SST can be skillfully predicted up to similar to 18 lead months when initialized on 1 November, but skill drops at similar to 12 lead months for May starts that encounter the boreal spring predictability barrier in year 2. The skill beyond the first year is highly conditioned to the phase of ENSO: Forecasts initialized at peak El Nino are more skillful in year 2 than those initialized at peak La Nina, with the transition to La Nina being more predictable than to El Nino. This asymmetry is related to the subsurface initial conditions in the western equatorial Pacific: peak El Nino states evolving into La Nina are associated with strong upperocean heat discharge of the western Pacific, the memory of which stays beyond 1 year. In contrast, the western Pacific recharged state associated with La Nina is usually weaker and shorter-lived, being a weaker preconditioner for subsequent El Nino, the year after. High prediction skill of ENSO events beyond 1 year provides motivation for extending the lead time of operational seasonal forecasts up to 2 years.