Using initial states to identify successful El Niño–Southern Oscillation forecasts 1 year in advance

El Niño-Southern Oscillation (ENSO) is the primary source of climate predictability on seasonal and longer timescales over much of the globe, but predicting its phase beyond 9 months has remained challenging. Here, we leverage an underappreciated factor in ENSO forecasting: the substantial variation in forecast skill based on the ENSO phase at the time a forecast is issued. Our analysis of transitions between ENSO phases in a dynamical hindcast dataset demonstrates that this variation markedly exceeds intermodel differences in predictive capability. Forecasts initialized during La Niña conditions perform best, with three models correctly predicting transitions to El Niño 1 year in advance over 70% of the time. We present our findings in a simple framework that allows decision-makers to assess the likelihood of a forecast being correct on the basis of the current ENSO state. Applied effectively, these findings could improve long-lead ENSO forecasting strategies, enhancing global risk management.