Predicting Atlantic and Benguela Niño events with deep learning

Abstract Extreme Atlantic and Benguela Niño events significantly impact the tropical Atlantic region with far-reaching consequences on local marine ecosystems¹,², African and South American climates³–⁵, to the El Niño Southern Oscillation⁶. While accurate forecasts of these events are invaluable, state-of-the-art dynamical seasonal forecasting systems have shown limited predictive capabilities⁷–⁹. Thus, the extent to which the variability of the tropical Atlantic is predictable is an open question. Here, exploiting a deep learning-based statistical prediction model, we show that tropical Atlantic events can be predicted up to 3-4 months in advance. Notably, our convolutional neural network model excels in forecasting peak-season events with remarkable accuracy extending lead-time up to 5 months. Detailed analysis reveals our model’s ability to exploit known physical precursors, particularly associated with long-wave ocean dynamics, for accurate predictions of Atlantic/Benguela Niños. This study challenges the perception that the tropical Atlantic is inherently unpredictable and highlights the potential of deep learning to advance our understanding and forecasting of critical climate events.