A deep learning method that can predict El Nino events up to one and a half years before they happen is described in Nature. The approach overcomes a long-standing challenge in the field of El Nino forecasting.
El Nino events originate in the eastern and central Pacific and can cause climate extremes and substantial damage to local ecosystems. Predicting these events has been difficult because conventional forecasting methods cannot provide an accurate prediction for lead times longer than one year.
Yoo-Geun Ham and colleagues present a deep learning approach that is able to predict El Nino events. The model was trained using historical climate data from 1871 to 1973 and simulations of El Nino events, and tested with data from 1984 to 2017. The deep learning algorithm was able to predict El Nino events with greater accuracy than current climate forecasts and with a longer lead time of up to one and a half years. The authors were also able to use their forecasting model to predict whether the event originated in the central or eastern Pacific, and identified sea surface temperature changes that precede an El Nino event.
The authors propose that the forecasts provided by this approach could also be used for future climate projections and to help inform policy responses to the impacts of El Nino.
Planetary science: Modelling electrolyte transport in water-rich exoplanetsNature Communications