Predicting Atlantic and Benguela Niño events with deep learning.

Area of Science:

• Climate Science
• Oceanography
• Artificial Intelligence

Background:

• Atlantic and Benguela Niño events significantly impact tropical Atlantic marine ecosystems, African climates, and the El Niño Southern Oscillation.
• Current dynamic forecasting systems have limited predictive skill for these crucial climate phenomena.
• The predictability of tropical Atlantic variability remains a significant scientific question.

Purpose of the Study:

• To investigate the potential of deep learning techniques for predicting Atlantic and Benguela Niño events.
• To assess the lead time and accuracy of deep learning models in forecasting these climate phenomena.
• To challenge the notion that the tropical Atlantic is inherently unpredictable.

Main Methods:

• Utilized a convolutional neural network (CNN) architecture, a type of deep learning model.
• Trained and evaluated the CNN on historical data to identify predictive patterns.
• Analyzed the model's ability to leverage known physical precursors for accurate forecasting.

Main Results:

• The deep learning model successfully predicted Atlantic/Benguela Niño events with lead times of 3 to 4 months.
• Peak-season events were forecasted with remarkable accuracy, extending the lead time to 5 months.
• The model demonstrated an ability to utilize physical precursors like long-wave ocean dynamics for prediction.

Conclusions:

• Deep learning offers a promising approach to significantly improve the prediction of Atlantic and Benguela Niño events.
• This study advances the understanding of tropical Atlantic variability and its predictability.
• The findings highlight the potential of AI in enhancing climate event forecasting and mitigating associated impacts.