A 2 Year Forecast for a 60–80% Chance of La Niña in 2017–2018

Historical observations show that one in two La Niña events last for two consecutive years. Despite their outsized impacts on drought, these 2 year La Niña are not predicted on a routine basis. Here we assess their predictability using retrospective forecasts performed with a climate model that simulates realistic multiyear events, as well as with an empirical model based on observed predictors. The skill of the retrospective forecasts allows us to make predictions for the upcoming 2017–2018 boreal winter starting from conditions in November 2015. These 2 year forecasts indicate that the return of La Niña is more likely than not, with a 60% probability based on the climate model and an 80% probability based on the empirical model; the likelihood of El Niño is less than 8% in both cases. These results demonstrate the feasibility of predictions of the duration of La Niña. Plain Language Summary Historical observations show that cold La Niña events in the tropical Pacific often reintensify for a second year. Despite their outsized climate impacts throughout the world, these 2 year La Niña events are not routinely predicted. Our study demonstrates that long‐term forecasts of these events are feasible. Moreover, we show increased likelihood of returning La Niña for next boreal winter, a result that is directly relevant for assessing climate risks throughout the world, for instance, over the southern tier of the U.S., where La Niña events create extreme seasonal heat and drought during winter and spring, and the Maritime Continent and Northern Australia, where it causes excess rainfall and flooding. Key Points Thermocline discharge driven by strong El Niño requires several years of La Niña conditions to return to a neutral state Forecast initialization at the peak of a strong El Niño event leads to skillful predictions of subsequent 2 year La Niña event Returning La Nina conditions in 2018 have 60% probability based on a dynamical model and 80% probability based on an empirical model